records for heap and btree. Minor change, mostly API changes to
pass through the required values. This is a simple change though
also provides the refactoring required for further enhancements
to conflict processing using the relOid. Changes only have effect
during Hot Standby.
restore_command, if the connection to the primary server is lost. This
ensures that the standby can recover automatically, if the connection is
lost for a long time and standby falls behind so much that the required
WAL segments have been archived and deleted in the master.
This also makes standby_mode useful without streaming replication; the
server will keep retrying restore_command every few seconds until the
trigger file is found. That's the same basic functionality pg_standby
offers, but without the bells and whistles.
To implement that, refactor the ReadRecord/FetchRecord functions. The
FetchRecord() function introduced in the original streaming replication
patch is removed, and all the retry logic is now in a new function called
XLogReadPage(). XLogReadPage() is now responsible for executing
restore_command, launching walreceiver, and waiting for new WAL to arrive
from primary, as required.
This also changes the life cycle of walreceiver. When launched, it now only
tries to connect to the master once, and exits if the connection fails, or
is lost during streaming for any reason. The startup process detects the
death, and re-launches walreceiver if necessary.
woken by alarm we send SIGUSR1 to all backends requesting that they
check to see if they are blocking Startup process. If so, they throw
ERROR/FATAL as for other conflict resolutions. Deadlock stop gap
removed. max_standby_delay = -1 option removed to prevent deadlock.
binary, revert PGDLLIMPORT decoration of global variables. I'm not sure
if there's any real harm from unnecessary PGDLLIMPORTs, but these are all
internal variables that external modules really shouldn't be messing
with. ThisTimeLineID still needs PGDLLIMPORT.
Conflict reason is passed through directly to the backend, so we can
take decisions about the effect of the conflict based upon the local
state. No specific changes, as yet, though this prepares for later work.
CancelVirtualTransaction() sends signals while holding ProcArrayLock.
Introduce errdetail_abort() to give message detail explaining that the
abort was caused by conflict processing. Remove CONFLICT_MODE states
in favour of using PROCSIG_RECOVERY_CONFLICT states directly, for clarity.
This includes two new kinds of postmaster processes, walsenders and
walreceiver. Walreceiver is responsible for connecting to the primary server
and streaming WAL to disk, while walsender runs in the primary server and
streams WAL from disk to the client.
Documentation still needs work, but the basics are there. We will probably
pull the replication section to a new chapter later on, as well as the
sections describing file-based replication. But let's do that as a separate
patch, so that it's easier to see what has been added/changed. This patch
also adds a new section to the chapter about FE/BE protocol, documenting the
protocol used by walsender/walreceivxer.
Bump catalog version because of two new functions,
pg_last_xlog_receive_location() and pg_last_xlog_replay_location(), for
monitoring the progress of replication.
Fujii Masao, with additional hacking by me
of this are to centralise the conflict code to allow further change,
as well as to allow passing through the full reason for the conflict
through to the conflicting backends. Backend state alters how we
can handle different types of conflict so this is now required.
As originally suggested by Heikki, no longer optional.
Previously we only cancelled sessions that were in-transaction.
Simple fix is to just cancel all sessions without waiting. Doing
it this way avoids complicating common code paths, which would
not be worth the trouble to cover this rare case.
Problem report and fix by Andres Freund, edited somewhat by me
deletion, so that we attempt to unlink the correct filepath. unlink()
errors are ignorable there, so lack of a DatabasePath initialization step
did not cause visible problems until a related bug showed up on Solaris.
Code refactored from xact_redo_commit() to
ProcessCommittedInvalidationMessages() in inval.c. Recovery may replay
shared invalidation messages for many databases, so we cannot
SetDatabasePath() once as we do in normal backends. Read the databaseid
from the shared invalidation messages, then set DatabasePath
temporarily before calling RelationCacheInitFileInvalidate().
Problem report by Robert Treat, analysis and fix by me.
This is more in keeping with modern practice, and is a first step towards
porting to Win64 (which has sizeof(pointer) > sizeof(long)).
Tsutomu Yamada, Magnus Hagander, Tom Lane
Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record.
New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far.
This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required.
Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit.
Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
This patch also removes buffer-usage statistics from the track_counts
output, since this (or the global server statistics) is deemed to be a better
interface to this information.
Itagaki Takahiro, reviewed by Euler Taveira de Oliveira.
via the "flat files" facility. This requires making it enough like a backend
to be able to run transactions; it's no longer an "auxiliary process" but
more like the autovacuum worker processes. Also, its signal handling has
to be brought into line with backends/workers. In particular, since it
now has to handle procsignal.c processing, the special autovac-launcher-only
signal conditions are moved to SIGUSR2.
Alvaro, with some cleanup from Tom
(That flat file is now completely useless, but removal will come later.)
To do this, postpone client authentication into the startup transaction
that's run by InitPostgres. We still collect the startup packet and do
SSL initialization (if needed) at the same time we did before. The
AuthenticationTimeout is applied separately to startup packet collection
and the actual authentication cycle. (This is a bit annoying, since it
means a couple extra syscalls; but the signal handling requirements inside
and outside a transaction are sufficiently different that it seems best
to treat the timeouts as completely independent.)
A small security disadvantage is that if the given database name is invalid,
this will be reported to the client before any authentication happens.
We could work around that by connecting to database "postgres" instead,
but consensus seems to be that it's not worth introducing such surprising
behavior.
Processing of all command-line switches and GUC options received from the
client is now postponed until after authentication. This means that
PostAuthDelay is much less useful than it used to be --- if you need to
investigate problems during InitPostgres you'll have to set PreAuthDelay
instead. However, allowing an unauthenticated user to set any GUC options
whatever seems a bit too risky, so we'll live with that.
To make this work in the base case, pg_database now has a nailed-in-cache
relation descriptor that is initialized using hardwired knowledge in
relcache.c. This means pg_database is added to the set of relations that
need to have a Schema_pg_xxx macro maintained in pg_attribute.h. When this
path is taken, we'll have to do a seqscan of pg_database to find the row
we need.
In the normal case, we are able to do an indexscan to find the database's row
by name. This is made possible by storing a global relcache init file that
describes only the shared catalogs and their indexes (and therefore is usable
by all backends in any database). A new backend loads this cache file,
finds its database OID after an indexscan on pg_database, and then loads
the local relcache init file for that database.
This change should effectively eliminate number of databases as a factor
in backend startup time, even with large numbers of databases. However,
the real reason for doing it is as a first step towards getting rid of
the flat files altogether. There are still several other sub-projects
to be tackled before that can happen.
fsync() fails, say "file" rather than "relation" when printing the filename.
This makes messages that display block numbers a bit confusing. For example,
in message 'could not read block 150000 of file "base/1234/5678.1"', 150000
is the block number from the beginning of the relation, ie. segment 0, not
150000th block within that segment. Per discussion, users aren't usually
interested in the exact location within the file, so we can live with that.
To ease constructing error messages, add FilePathName(File) function to
return the pathname of a virtual fd.
This patch gets us out from under the Unix limitation of two user-defined
signal types. We already had done something similar for signals directed to
the postmaster process; this adds multiplexing for signals directed to
backends and auxiliary processes (so long as they're connected to shared
memory).
As proof of concept, replace the former usage of SIGUSR1 and SIGUSR2
for backends with use of the multiplexing mechanism. There are still some
hard-wired definitions of SIGUSR1 and SIGUSR2 for other process types,
but getting rid of those doesn't seem interesting at the moment.
Fujii Masao
After a patch originally submitted by Nobuhiro Iwamatsu, but corrected
(I think) to match our guidelines for safe use of asm fragments.
This should be considered untested ...
during it:
When bgwriter is active, the startup process can't perform mdsync() correctly
because it won't see the fsync requests accumulated in bgwriter's private
pendingOpsTable. Therefore make bgwriter responsible for the end-of-recovery
checkpoint as well, when it's active.
When bgwriter is active (= archive recovery), the startup process must not
accumulate fsync requests to its own pendingOpsTable, since bgwriter won't
see them there when it performs restartpoints. Make startup process drop its
pendingOpsTable when bgwriter is launched to avoid that.
Update minimum recovery point one last time when leaving archive recovery.
It won't be updated by the end-of-recovery checkpoint because XLogFlush()
sees us as out of recovery already.
This fixes bug #4879 reported by Fujii Masao.
a backend has done exit(0) or exit(1) without having disengaged itself
from shared memory. We are at risk for this whenever third-party code is
loaded into a backend, since such code might not know it's supposed to go
through proc_exit() instead. Also, it is reported that under Windows
there are ways to externally kill a process that cause the status code
returned to the postmaster to be indistinguishable from a voluntary exit
(thank you, Microsoft). If this does happen then the system is probably
hosed --- for instance, the dead session might still be holding locks.
So the best recovery method is to treat this like a backend crash.
The dead man switch is armed for a particular child process when it
acquires a regular PGPROC, and disarmed when the PGPROC is released;
these should be the first and last touches of shared memory resources
in a backend, or close enough anyway. This choice means there is no
coverage for auxiliary processes, but I doubt we need that, since they
shouldn't be executing any user-provided code anyway.
This patch also improves the management of the EXEC_BACKEND
ShmemBackendArray array a bit, by reducing search costs.
Although this problem is of long standing, the lack of field complaints
seems to mean it's not critical enough to risk back-patching; at least
not till we get some more testing of this mechanism.
when we are waiting for old snapshots to go away during a concurrent index
build. In particular, this rule lets us avoid waiting for
idle-in-transaction sessions.
This logic could be improved further if we had some way to wake up when
the session we are currently waiting for goes idle-in-transaction. However
that would be a significantly more complex/invasive patch, so it'll have to
wait for some other day.
Simon Riggs, with some improvements by Tom.
As pointed out by ITAGAKI Takahiro, we split SInvalLock into two in 8.4,
so to keep the numbers of the rest of the locks unchanged from 8.3, we
don't need a placeholder.
its usual buffer cleaning duties during archive recovery, and it's responsible
for performing restartpoints.
This requires some changes in postmaster. When the startup process has done
all the initialization and is ready to start WAL redo, it signals the
postmaster to launch the background writer. The postmaster is signaled again
when the point in recovery is reached where we know that the database is in
consistent state. Postmaster isn't interested in that at the moment, but
that's the point where we could let other backends in to perform read-only
queries. The postmaster is signaled third time when the recovery has ended,
so that postmaster knows that it's safe to start accepting connections.
The startup process now traps SIGTERM, and performs a "clean" shutdown. If
you do a fast shutdown during recovery, a shutdown restartpoint is performed,
like a shutdown checkpoint, and postmaster kills the processes cleanly. You
still have to continue the recovery at next startup, though.
Currently, the background writer is only launched during archive recovery.
We could launch it during crash recovery as well, but it seems better to keep
that codepath as simple as possible, for the sake of robustness. And it
couldn't do any restartpoints during crash recovery anyway, so it wouldn't be
that useful.
log_restartpoints is gone. Use log_checkpoints instead. This is yet to be
documented.
This whole operation is a pre-requisite for Hot Standby, but has some value of
its own whether the hot standby patch makes 8.4 or not.
Simon Riggs, with lots of modifications by me.
GUC variable effective_io_concurrency controls how many concurrent block
prefetch requests will be issued.
(The best way to handle this for plain index scans is still under debate,
so that part is not applied yet --- tgl)
Greg Stark
initialization, to give loadable modules a reasonable place to perform
creation of any shared memory areas they need. This is the logical conclusion
of our previous creation of RequestAddinShmemSpace() and RequestAddinLWLocks().
We don't need an explicit shmem_shutdown_hook, because the existing
on_shmem_exit and on_proc_exit mechanisms serve that need.
Also, adjust SubPostmasterMain so that libraries that got loaded into the
postmaster will be loaded into all child processes, not only regular backends.
This improves consistency with the non-EXEC_BACKEND behavior, and might be
necessary for functionality for some types of add-ons.
replication patch needs a signal, but we've already used SIGUSR1 and
SIGUSR2 in normal backends. This patch allows reusing SIGUSR1 for that,
and for other purposes too if the need arises.
heap page, where a set bit indicates that all tuples on the page are
visible to all transactions, and the page therefore doesn't need
vacuuming. It is stored in a new relation fork.
Lazy vacuum uses the visibility map to skip pages that don't need
vacuuming. Vacuum is also responsible for setting the bits in the map.
In the future, this can hopefully be used to implement index-only-scans,
but we can't currently guarantee that the visibility map is always 100%
up-to-date.
In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on
each heap page, also indicating that all tuples on the page are visible to
all transactions. It's important that this flag is kept up-to-date. It
is also used to skip visibility tests in sequential scans, which gives a
small performance gain on seqscans.
truncations in FSM code, call FreeSpaceMapTruncateRel from smgr_redo. To
make that cleaner from modularity point of view, move the WAL-logging one
level up to RelationTruncate, and move RelationTruncate and all the
related WAL-logging to new src/backend/catalog/storage.c file. Introduce
new RelationCreateStorage and RelationDropStorage functions that are used
instead of calling smgrcreate/smgrscheduleunlink directly. Move the
pending rel deletion stuff from smgrcreate/smgrscheduleunlink to the new
functions. This leaves smgr.c as a thin wrapper around md.c; all the
transactional stuff is now in storage.c.
This will make it easier to add new forks with similar truncation logic,
like the visibility map.
(but not locked, as that would risk deadlocks). Also, make it work in a small
ring of buffers to avoid having bulk inserts trash the whole buffer arena.
Robert Haas, after an idea of Simon Riggs'.
allowed different processes to have different addresses for the shmem segment
in quite a long time, but there were still a few places left that used the
old coding convention. Clean them up to reduce confusion and improve the
compiler's ability to detect pointer type mismatches.
Kris Jurka
on non-full-page-image WAL records, and quite arbitrarily, only if there's
less than 20% free space on the page after the insert/update (not on HOT
updates, though). The 20% cutoff should avoid most of the overhead, when
replaying a bulk insertion, for example, while ensuring that pages that
are full are marked as full in the FSM.
This is mostly to avoid the nasty worst case scenario, where you replay
from a PITR archive, and the FSM information in the base backup is really
out of date. If there was a lot of pages that the outdated FSM claims to
have free space, but don't actually have any, the first unlucky inserter
after the recovery would traverse through all those pages, just to find
out that they're full. We didn't have this problem with the old FSM
implementation, because we simply threw the FSM information away on a
non-clean shutdown.
functions into one ReadBufferExtended function, that takes the strategy
and mode as argument. There's three modes, RBM_NORMAL which is the default
used by plain ReadBuffer(), RBM_ZERO, which replaces ZeroOrReadBuffer, and
a new mode RBM_ZERO_ON_ERROR, which allows callers to read corrupt pages
without throwing an error. The FSM needs the new mode to recover from
corrupt pages, which could happend if we crash after extending an FSM file,
and the new page is "torn".
Add fork number to some error messages in bufmgr.c, that still lacked it.
This basically takes some build system code that was previously labeled
"Solaris" and ties it to the compiler rather than the operating system.
Author: Julius Stroffek <Julius.Stroffek@Sun.COM>
relation forks. While the file names are not visible to users, for those
that do peek into the data directory, it's nice to have more descriptive
names. Per Greg Stark's suggestion.
name of a fork ('main' or 'fsm', at the moment) to pg_relation_size() to
get the size of a specific fork. Defaults to 'main', if none given.
While we're at it, modify pg_relation_size to take a regclass as argument,
instead of separate variants taking oid and name. This change is
transparent to typical use where the table name is passed as a string
literal, like pg_relation_size('table'), but will break queries like
pg_relation_size(namecol), where namecol is of type name. text-type input
still works, and using a non-schema-qualified table name is not very
reliable anyway, so this is unlikely to break anyone's queries in practice.
free space information is stored in a dedicated FSM relation fork, with each
relation (except for hash indexes; they don't use FSM).
This eliminates the max_fsm_relations and max_fsm_pages GUC options; remove any
trace of them from the backend, initdb, and documentation.
Rewrite contrib/pg_freespacemap to match the new FSM implementation. Also
introduce a new variant of the get_raw_page(regclass, int4, int4) function in
contrib/pageinspect that let's you to return pages from any relation fork, and
a new fsm_page_contents() function to inspect the new FSM pages.
forestalls potential overflow when the same table (or other object, but
usually tables) is accessed by very many successive queries within a single
transaction. Per report from Michael Milligan.
Back-patch to 8.0, which is as far back as the patch conveniently applies.
There have been no reports of overflow in pre-8.3 releases, but clearly the
risk existed all along. (Michael's report suggests that 8.3 may consume lock
counts faster than prior releases, but with no test case to look at it's hard
to be sure about that. Widening the counts seems a good future-proofing
measure in any event.)
of multiple forks, and each fork can be created and grown separately.
The bulk of this patch is about changing the smgr API to include an extra
ForkNumber argument in every smgr function. Also, smgrscheduleunlink and
smgrdounlink no longer implicitly call smgrclose, because other forks might
still exist after unlinking one. The callers of those functions have been
modified to call smgrclose instead.
This patch in itself doesn't have any user-visible effect, but provides the
infrastructure needed for upcoming patches. The additional forks envisioned
are a rewritten FSM implementation that doesn't rely on a fixed-size shared
memory block, and a visibility map to allow skipping portions of a table in
VACUUM that have no dead tuples.
or target database is being accessed by other users, it tells you whether
the "other users" are live sessions or uncommitted prepared transactions.
(Indeed, it tells you exactly how many of each, but that's mostly just
because it was easy to do so.) This should help forestall the gotcha of
not realizing that a prepared transaction is what's blocking the command.
Per discussion.
macros patch :-(. Results from both baiji and mastodon imply that MSVC
fails to perceive offsetof(PageHeaderData, pd_linp[0]) as a constant
expression in some contexts where offsetof(PageHeaderData, pd_linp) works
fine. Sloth, thy name is Micro.
thereby forestalling any problems with alignment of the data structure placed
there. Since SizeOfPageHeaderData is maxalign'd anyway in 8.3 and HEAD, this
does not actually change anything right now, but it is foreseeable that the
header size will change again someday. I had to fix a couple of places that
were assuming that the content offset is just SizeOfPageHeaderData rather than
MAXALIGN(SizeOfPageHeaderData). Per discussion of Zdenek's page-macros patch.
backend. If so, send a LOG message to the postmaster log, and if the table
is beyond the vacuum-for-wraparound horizon, forcibly drop it. Per recent
discussions. Perhaps we ought to back-patch this, but it probably needs
to age a bit in HEAD first.
unnecessary cache resets. The major changes are:
* When the queue overflows, we only issue a cache reset to the specific
backend or backends that still haven't read the oldest message, rather
than resetting everyone as in the original coding.
* When we observe backend(s) falling well behind, we signal SIGUSR1
to only one backend, the one that is furthest behind and doesn't already
have a signal outstanding for it. When it finishes catching up, it will
in turn signal SIGUSR1 to the next-furthest-back guy, if there is one that
is far enough behind to justify a signal. The PMSIGNAL_WAKEN_CHILDREN
mechanism is removed.
* We don't attempt to clean out dead messages after every message-receipt
operation; rather, we do it on the insertion side, and only when the queue
fullness passes certain thresholds.
* Split SInvalLock into SInvalReadLock and SInvalWriteLock so that readers
don't block writers nor vice versa (except during the infrequent queue
cleanout operations).
* Transfer multiple sinval messages for each acquisition of a read or
write lock.
corresponding struct definitions. This allows other headers to avoid including
certain highly-loaded headers such as rel.h and relscan.h, instead using just
relcache.h, heapam.h or genam.h, which are more lightweight and thus cause less
unnecessary dependencies.
forks. XLogOpenRelation() and the associated light-weight relation cache in
xlogutils.c is gone, and XLogReadBuffer() now takes a RelFileNode as argument,
instead of Relation.
For functions that still need a Relation struct during WAL replay, there's a
new function called CreateFakeRelcacheEntry() that returns a fake entry like
XLogOpenRelation() used to.
more logical that way, and also it reduces the amount of unnecessary includes
in bufpage.h, which is widely used.
Zdenek Kotala.
My previous patch to bufpage.h should also have credited him as author, but I
forgot (sorry about that).
There are two ways to track a snapshot: there's the "registered" list, which
is used for arbitrary long-lived snapshots; and there's the "active stack",
which is used for the snapshot that is considered "active" at any time.
This also allows users of snapshots to stop worrying about snapshot memory
allocation and freeing, and about using PG_TRY blocks around ActiveSnapshot
assignment. This is all done automatically now.
As a consequence, this allows us to reset MyProc->xmin when there are no
more snapshots registered in the current backend, reducing the impact that
long-running transactions have on VACUUM.
unnecessary #include lines in it. Also, move some tuple routine prototypes and
macros to htup.h, which allows removal of heapam.h inclusion from some .c
files.
For this to work, a new header file access/sysattr.h needed to be created,
initially containing attribute numbers of system columns, for pg_dump usage.
While at it, make contrib ltree, intarray and hstore header files more
consistent with our header style.
corrupted. (Neither is very important if SIGTERM is used to shut down the
whole database cluster together, but there's a problem if someone tries to
SIGTERM individual backends.) To do this, introduce new infrastructure
macros PG_ENSURE_ERROR_CLEANUP/PG_END_ENSURE_ERROR_CLEANUP that take care
of transiently pushing an on_shmem_exit cleanup hook. Also use this method
for createdb cleanup --- that wasn't a shared-memory-corruption problem,
but SIGTERM abort of createdb could leave orphaned files lying around.
Backpatch as far as 8.2. The shmem corruption cases don't exist in 8.1,
and the createdb usage doesn't seem important enough to risk backpatching
further.
snapmgmt.c file for the former. The header files have also been reorganized
in three parts: the most basic snapshot definitions are now in a new file
snapshot.h, and the also new snapmgmt.h keeps the definitions for snapmgmt.c.
tqual.h has been reduced to the bare minimum.
This patch is just a first step towards managing live snapshots within a
transaction; there is no functionality change.
Per my proposal to pgsql-patches on 20080318191940.GB27458@alvh.no-ip.org and
subsequent discussion.
deals with the queue, including locking etc, is all in sinvaladt.c. This means
that the struct definition of the queue, and the queue pointer, are now
internal "implementation details" inside sinvaladt.c.
Per my proposal dated 25-Jun-2007 and followup discussion.
than dividing them into 1GB segments as has been our longtime practice. This
requires working support for large files in the operating system; at least for
the time being, it won't be the default.
Zdenek Kotala
temporary table; we can't support that because there's no way to clean up the
source backend's internal state if the eventual COMMIT PREPARED is done by
another backend. This was checked correctly in 8.1 but I broke it in 8.2 :-(.
Patch by Heikki Linnakangas, original trouble report by John Smith.
whether to execute an immediate interrupt, rather than testing whether
LockWaitCancel() cancelled a lock wait. The old way misclassified the case
where we were blocked in ProcWaitForSignal(), and arguably would misclassify
any other future additions of new ImmediateInterruptOK states too. This
allows reverting the old kluge that gave LockWaitCancel() a return value,
since no callers care anymore. Improve comments in the various
implementations of PGSemaphoreLock() to explain that on some platforms, the
assumption that semop() exits after a signal is wrong, and so we must ensure
that the signal handler itself throws elog if we want cancel or die interrupts
to be effective. Per testing related to bug #3883, though this patch doesn't
solve those problems fully.
Perhaps this change should be back-patched, but since pre-8.3 branches aren't
really relying on autovacuum to respond to SIGINT, it doesn't seem critical
for them.
checkpoint. This guards against an unlikely data-loss scenario in which
we re-use the relfilenode, then crash, then replay the deletion and
recreation of the file. Even then we'd be OK if all insertions into the
new relation had been WAL-logged ... but that's not guaranteed given all
the no-WAL-logging optimizations that have recently been added.
Patch by Heikki Linnakangas, per a discussion last month.
having several of them. Add two more flags: whether the process is
executing an ANALYZE, and whether a vacuum is for Xid wraparound (which
is obviously only set by autovacuum).
Sneakily move the worker's recently-acquired PostAuthDelay to a more useful
place.
buffers that cannot possibly need to be cleaned, and estimates how many
buffers it should try to clean based on moving averages of recent allocation
requests and density of reusable buffers. The patch also adds a couple
more columns to pg_stat_bgwriter to help measure the effectiveness of the
bgwriter.
Greg Smith, building on his own work and ideas from several other people,
in particular a much older patch from Itagaki Takahiro.
unpruned XMAX in its header. At the cost of 4 bytes per page, this keeps us
from performing heap_page_prune when there's no chance of pruning anything.
Seems to be necessary per Heikki's preliminary performance testing.
columns, and the new version can be stored on the same heap page, we no longer
generate extra index entries for the new version. Instead, index searches
follow the HOT-chain links to ensure they find the correct tuple version.
In addition, this patch introduces the ability to "prune" dead tuples on a
per-page basis, without having to do a complete VACUUM pass to recover space.
VACUUM is still needed to clean up dead index entries, however.
Pavan Deolasee, with help from a bunch of other people.
than two independent bits (one of which was never used in heap pages anyway,
or at least hadn't been in a very long time). This gives us flexibility to
add the HOT notions of redirected and dead item pointers without requiring
anything so klugy as magic values of lp_off and lp_len. The state values
are chosen so that for the states currently in use (pre-HOT) there is no
change in the physical representation.
ReadNewTransactionId from GetSnapshotData --- with a "latestCompletedXid"
variable that is updated during transaction commit or abort. Since
latestCompletedXid is written only in places that had to lock ProcArrayLock
exclusively anyway, and is read only in places that had to lock ProcArrayLock
shared anyway, it adds no new locking requirements to the system despite being
cluster-wide. Moreover, removing ReadNewTransactionId from snapshot
acquisition eliminates the need to take both XidGenLock and ProcArrayLock at
the same time. Since XidGenLock is sometimes held across I/O this can be a
significant win. Some preliminary benchmarking suggested that this patch has
no effect on average throughput but can significantly improve the worst-case
transaction times seen in pgbench. Concept by Florian Pflug, implementation
by Tom Lane.
rows will normally never obtain an XID at all. We already did things this way
for subtransactions, but this patch extends the concept to top-level
transactions. In applications where there are lots of short read-only
transactions, this should improve performance noticeably; not so much from
removal of the actual XID-assignments, as from reduction of overhead that's
driven by the rate of XID consumption. We add a concept of a "virtual
transaction ID" so that active transactions can be uniquely identified even
if they don't have a regular XID. This is a much lighter-weight concept:
uniqueness of VXIDs is only guaranteed over the short term, and no on-disk
record is made about them.
Florian Pflug, with some editorialization by Tom.
with the recent patch to log temp file sizes at removal time. Doesn't seem
worth fixing since it's unused.
In passing, make a few elog messages conform to the message style guide.
over a fairly long period of time, rather than being spat out in a burst.
This happens only for background checkpoints carried out by the bgwriter;
other cases, such as a shutdown checkpoint, are still done at full speed.
Remove the "all buffers" scan in the bgwriter, and associated stats
infrastructure, since this seems no longer very useful when the checkpoint
itself is properly throttled.
Original patch by Itagaki Takahiro, reworked by Heikki Linnakangas,
and some minor API editorialization by me.
within a signal handler (this might be safe given the relatively narrow code
range in which the interrupt is enabled, but it seems awfully risky); do issue
more informative log messages that tell what is being waited for and the exact
length of the wait; minor other code cleanup. Greg Stark and Tom Lane
for each temp file, rather than once per sort or hashjoin; this allows
spreading the data of a large sort or join across multiple tablespaces.
(I remain dubious that this will make any difference in practice, but certain
people insisted.) Arrange to cache the results of parsing the GUC variable
instead of recomputing from scratch on every demand, and push usage of the
cache down to the bottommost fd.c level.
tablespace(s) in which to store temp tables and temporary files. This is a
list to allow spreading the load across multiple tablespaces (a random list
element is chosen each time a temp object is to be created). Temp files are
not stored in per-database pgsql_tmp/ directories anymore, but per-tablespace
directories.
Jaime Casanova and Albert Cervera, with review by Bernd Helmle and Tom Lane.
will exit before failing because of conflicting DB usage. Per discussion,
this seems a good idea to help mask the fact that backend exit takes nonzero
time. Remove a couple of thereby-obsoleted sleeps in contrib and PL
regression test sequences.
buffers, rather than blowing out the whole shared-buffer arena. Aside from
avoiding cache spoliation, this fixes the problem that VACUUM formerly tended
to cause a WAL flush for every page it modified, because we had it hacked to
use only a single buffer. Those flushes will now occur only once per
ring-ful. The exact ring size, and the threshold for seqscans to switch into
the ring usage pattern, remain under debate; but the infrastructure seems
done. The key bit of infrastructure is a new optional BufferAccessStrategy
object that can be passed to ReadBuffer operations; this replaces the former
StrategyHintVacuum API.
This patch also changes the buffer usage-count methodology a bit: we now
advance usage_count when first pinning a buffer, rather than when last
unpinning it. To preserve the behavior that a buffer's lifetime starts to
decrease when it's released, the clock sweep code is modified to not decrement
usage_count of pinned buffers.
Work not done in this commit: teach GiST and GIN indexes to use the vacuum
BufferAccessStrategy for vacuum-driven fetches.
Original patch by Simon, reworked by Heikki and again by Tom.
from the WAL data, don't bother to physically read it; just have bufmgr.c
return a zeroed-out buffer instead. This speeds recovery significantly,
and also avoids unnecessary failures when a page-to-be-overwritten has corrupt
page headers on disk. This replaces a former kluge that accomplished the
latter by pretending zero_damaged_pages was always ON during WAL recovery;
which was OK when the kluge was put in, but is unsafe when restoring a WAL
log that was written with full_page_writes off.
Heikki Linnakangas
processes to be running simultaneously. Also, now autovacuum processes do not
count towards the max_connections limit; they are counted separately from
regular processes, and are limited by the new GUC variable
autovacuum_max_workers.
The launcher now has intelligence to launch workers on each database every
autovacuum_naptime seconds, limited only on the max amount of worker slots
available.
Also, the global worker I/O utilization is limited by the vacuum cost-based
delay feature. Workers are "balanced" so that the total I/O consumption does
not exceed the established limit. This part of the patch was contributed by
ITAGAKI Takahiro.
Per discussion.
are in their commit critical sections via flags in the ProcArray. Checkpoint
can watch the ProcArray to determine when it's safe to proceed. This is
a considerably better solution to the original problem of race conditions
between checkpoint and transaction commit: it speeds up commit, since there's
one less lock to fool with, and it prevents the problem of checkpoint being
delayed indefinitely when there's a constant flow of commits. Heikki, with
some kibitzing from Tom.
this, add a 16-bit "flags" field to page headers by stealing some bits from
pd_tli. We use one flag bit as a hint to indicate whether there are any
unused line pointers; the remaining 15 are available for future use.
This is a cut-down form of an idea proposed by Hiroki Kataoka in July 2005.
At the time it was rejected because the original patch increased the size of
page headers and it wasn't clear that the benefit outweighed the distributed
cost. The flag-bit approach gets most of the benefit without requiring an
increase in the page header size.
Heikki Linnakangas and Tom Lane
I refactored findsplitloc and checksplitloc so that the division of
labor is more clear IMO. I pushed all the space calculation inside the
loop to checksplitloc.
I also fixed the off by 4 in free space calculation caused by
PageGetFreeSpace subtracting sizeof(ItemIdData), even though it was
harmless, because it was distracting and I felt it might come back to
bite us in the future if we change the page layout or alignments.
There's now a new function PageGetExactFreeSpace that doesn't do the
subtraction.
findsplitloc now tries the "just the new item to right page" split as
well. If people don't like the refactoring, I can write a patch to just
add that.
Heikki Linnakangas
continuously, and requests vacuum runs of "autovacuum workers" to postmaster.
The workers do the actual vacuum work. This allows for future improvements,
like allowing multiple autovacuum jobs running in parallel.
For now, the code keeps the original behavior of having a single autovac
process at any time by sleeping until the previous worker has finished.
keeping private state in each backend that has inserted and deleted the same
tuple during its current top-level transaction. This is sufficient since
there is no need to be able to determine the cmin/cmax from any other
transaction. This gets us back down to 23-byte headers, removing a penalty
paid in 8.0 to support subtransactions. Patch by Heikki Linnakangas, with
minor revisions by moi, following a design hashed out awhile back on the
pghackers list.
accessing it, like DROP DATABASE. This allows the regression tests to pass
with autovacuum enabled, which open the gates for finally enabling autovacuum
by default.
having md.c return a success/failure boolean to smgr.c, which was just going
to elog anyway, let md.c issue the elog messages itself. This allows better
error reporting, particularly in cases such as "short read" or "short write"
which Peter was complaining of. Also, remove the kluge of allowing mdread()
to return zeroes from a read-beyond-EOF: this is now an error condition
except when InRecovery or zero_damaged_pages = true. (Hash indexes used to
require that behavior, but no more.) Also, enforce that mdwrite() is to be
used for rewriting existing blocks while mdextend() is to be used for
extending the relation EOF. This restriction lets us get rid of the old
ad-hoc defense against creating huge files by an accidental reference to
a bogus block number: we'll only create new segments in mdextend() not
mdwrite() or mdread(). (Again, when InRecovery we allow it anyway, since
we need to allow updates of blocks that were later truncated away.)
Also, clean up the original makeshift patch for bug #2737: move the
responsibility for padding relation segments to full length into md.c.
modules; the first try was not usable in EXEC_BACKEND builds (e.g.,
Windows). Instead, just provide some entry points to increase the
allocation requests during postmaster start, and provide a dedicated
LWLock that can be used to synchronize allocation operations performed
by backends. Per discussion with Marc Munro.
even when a single relation requires more than max_fsm_pages pages. Also,
make VACUUM emit a warning in this case, since it likely means that VACUUM
FULL or other drastic corrective measure is needed. Per reports from Jeff
Frost and others of unexpected changes in the claimed max_fsm_pages need.
contrib functionality. Along the way, remove the USER_LOCKS configuration
symbol, since it no longer makes any sense to try to compile that out.
No user documentation yet ... mmoncure has promised to write some.
Thanks to Abhijit Menon-Sen for creating a first draft to work from.
locks that would conflict with a specified lock request, without
actually trying to get that lock. Use this instead of the former ad hoc
method of doing the first wait step in CREATE INDEX CONCURRENTLY.
Fixes problem with undetected deadlock and in many cases will allow the
index creation to proceed sooner than it otherwise could've. Per
discussion with Greg Stark.
the rel, it's easy to get rid of the narrow race-condition window that
used to exist in VACUUM and CLUSTER. Did some minor code-beautification
work in the same area, too.
(table or index) before trying to open its relcache entry. This fixes
race conditions in which someone else commits a change to the relation's
catalog entries while we are in process of doing relcache load. Problems
of that ilk have been reported sporadically for years, but it was not
really practical to fix until recently --- for instance, the recent
addition of WAL-log support for in-place updates helped.
Along the way, remove pg_am.amconcurrent: all AMs are now expected to support
concurrent update.
vacuums. This allows a OLTP-like system with big tables to continue
regular vacuuming on small-but-frequently-updated tables while the
big tables are being vacuumed.
Original patch from Hannu Krossing, rewritten by Tom Lane and updated
by me.
To this end, add a couple of columns to pg_class, relminxid and relvacuumxid,
based on which we calculate the pg_database columns after each vacuum.
We now force all databases to be vacuumed, even template ones. A backend
noticing too old a database (meaning pg_database.datminxid is in danger of
falling behind Xid wraparound) will signal the postmaster, which in turn will
start an autovacuum iteration to process the offending database. In principle
this is only there to cope with frozen (non-connectable) databases without
forcing users to set them to connectable, but it could force regular user
database to go through a database-wide vacuum at any time. Maybe we should
warn users about this somehow. Of course the real solution will be to use
autovacuum all the time ;-)
There are some additional improvements we could have in this area: for example
the vacuum code could be smarter about not updating pg_database for each table
when called by autovacuum, and do it only once the whole autovacuum iteration
is done.
I updated the system catalogs documentation, but I didn't modify the
maintenance section. Also having some regression tests for this would be nice
but it's not really a very straightforward thing to do.
Catalog version bumped due to system catalog changes.
current commands; instead, store current-status information in shared
memory. This substantially reduces the overhead of stats_command_string
and also ensures that pg_stat_activity is fully up to date at all times.
Per my recent proposal.
o remove many WIN32_CLIENT_ONLY defines
o add WIN32_ONLY_COMPILER define
o add 3rd argument to open() for portability
o add include/port/win32_msvc directory for
system includes
Magnus Hagander
into a single mostly-physical-order scan of the index. This requires some
ticklish interlocking considerations, but should create no material
performance impact on normal index operations (at least given the
already-committed changes to make scans work a page at a time). VACUUM
itself should get significantly faster in any index that's degenerated to a
very nonlinear page order. Also, we save one pass over the index entirely,
except in the case where there were no deletions to do and so only one pass
happened anyway.
Original patch by Heikki Linnakangas, rework by Tom Lane.
set to the large object context ("fscxt"), as this is inevitably a source of
transaction-duration memory leaks. Not sure why we'd not noticed it before;
maybe people weren't touching a whole lot of LOs in the same transaction
before the 8.1 pg_dump changes. Per report from Wayne Conrad.
Backpatched as far as 8.1, but the problem doubtless goes all the way back.
I'm disinclined to spend the time to try to verify that the older branches
would still work if patched, seeing that this code was significantly modified
for 8.0 and again for 8.1, and that we don't have any trouble reports before
8.1. (Maybe the leaks were smaller before?)
symbol for PPC64 hardware. I hadn't known that Apple supported PPC64 at
all, but darn if there aren't 64-bit variant libraries in OS X as well
as support in their gcc.
If a process abandons a wait in LockBufferForCleanup (in practice,
only happens if someone cancels a VACUUM) just before someone else
sends it a signal indicating the buffer is available, it was possible
for the wakeup to remain in the process' semaphore, causing misbehavior
next time the process waited for an lmgr lock. Rather than try to
prevent the race condition directly, it seems best to make the lock
manager robust against leftover wakeups, by having it repeat waiting
on the semaphore if the lock has not actually been granted or denied
yet.
misleadingly-named WriteBuffer routine, and instead require routines that
change buffer pages to call MarkBufferDirty (which does exactly what it says).
We also require that they do so before calling XLogInsert; this takes care of
the synchronization requirement documented in SyncOneBuffer. Note that
because bufmgr takes the buffer content lock (in shared mode) while writing
out any buffer, it doesn't matter whether MarkBufferDirty is executed before
the buffer content change is complete, so long as the content change is
completed before releasing exclusive lock on the buffer. So it's OK to set
the dirtybit before we fill in the LSN.
This eliminates the former kluge of needing to set the dirtybit in LockBuffer.
Aside from making the code more transparent, we can also add some new
debugging assertions, in particular that the caller of MarkBufferDirty must
hold the buffer content lock, not merely a pin.
This commit doesn't make much functional change, but it does eliminate some
duplicated code --- for instance, PageIsNew tests are now done inside
XLogReadBuffer rather than by each caller.
The GIST xlog code still needs a lot of love, but I'll worry about that
separately.
when an error occurs during xlog replay. Also, replace the former risky
'write into a fixed-size buffer with no overflow detection' API for XLOG
record description routines; use an expansible StringInfo instead. (The
latter accounts for most of the patch bulk.)
Qingqing Zhou
occurs when it tries to heap_open pg_tablespace. When control returns to
smgrcreate, that routine will be holding a dangling pointer to a closed
SMgrRelation, resulting in mayhem. This is of course a consequence of
the violation of proper module layering inherent in having smgr.c call
a tablespace command routine, but the simplest fix seems to be to change
the locking mechanism. There's no real need for TablespaceCreateDbspace
to touch pg_tablespace at all --- it's only opening it as a way of locking
against a parallel DROP TABLESPACE command. A much better answer is to
create a special-purpose LWLock to interlock these two operations.
This drops TablespaceCreateDbspace quite a few layers down the food chain
and makes it something reasonably safe for smgr to call.
an LWLock instead of a spinlock. This hardly matters on Unix machines
but should improve startup performance on Windows (or any port using
EXEC_BACKEND). Per previous discussion.
in favor of having just one set of macros that don't do HOLD/RESUME_INTERRUPTS
(hence, these correspond to the old SpinLockAcquire_NoHoldoff case).
Given our coding rules for spinlock use, there is no reason to allow
CHECK_FOR_INTERRUPTS to be done while holding a spinlock, and also there
is no situation where ImmediateInterruptOK will be true while holding a
spinlock. Therefore doing HOLD/RESUME_INTERRUPTS while taking/releasing a
spinlock is just a waste of cycles. Qingqing Zhou and Tom Lane.
file. The original code probed the PGPROC array separately for each PID,
which was not good for large numbers of backends: not only is the runtime
O(N^2) but most of it is spent holding ProcArrayLock. Instead, take the
lock just once and copy the active PIDs into an array, then use qsort
and bsearch so that the lookup time is more like O(N log N).
reduce contention for the former single LockMgrLock. Per my recent
proposal. I set it up for 16 partitions, but on a pgbench test this
gives only a marginal further improvement over 4 partitions --- we need
to test more scenarios to choose the number of partitions.
the data defining the semantics of a lock method (ie, conflict resolution
table and ancillary data, which is all constant) and the hash tables
storing the current state. The only thing we give up by this is the
ability to use separate hashtables for different lock methods, but there
is no need for that anyway. Put some extra fields into the LockMethod
definition structs to clean up some other uglinesses, like hard-wired
tests for DEFAULT_LOCKMETHOD and USER_LOCKMETHOD. This commit doesn't
do anything about the performance issues we were discussing, but it clears
away some of the underbrush that's in the way of fixing that.
comment line where output as too long, and update typedefs for /lib
directory. Also fix case where identifiers were used as variable names
in the backend, but as typedefs in ecpg (favor the backend for
indenting).
Backpatch to 8.1.X.
that was added to localbuf.c in 8.1; therefore, applying it to a temp table
left corrupt lookup state in memory. The only case where this had a
significant chance of causing problems was an ON COMMIT DELETE ROWS temp
table; the other possible paths left bogus state that was unlikely to
be used again. Per report from Csaba Nagy.
pointers, to ensure that compilers won't rearrange accesses to occur
while we're not holding the buffer header spinlock. It's probably
not necessary to mark volatile in every single place in bufmgr.c,
but better safe than sorry. Per trouble report from Kevin Grittner.
whether we seem to be running in a uniprocessor or multiprocessor.
The adjustment rules could probably still use further tweaking, but
I'm convinced this should be a win overall.
to 'Size' (that is, size_t), and install overflow detection checks in it.
This allows us to remove the former arbitrary restrictions on NBuffers
etc. It won't make any difference in a 32-bit machine, but in a 64-bit
machine you could theoretically have terabytes of shared buffers.
(How efficiently we could manage 'em remains to be seen.) Similarly,
num_temp_buffers, work_mem, and maintenance_work_mem can be set above
2Gb on a 64-bit machine. Original patch from Koichi Suzuki, additional
work by moi.
computation. On modern machines this is as fast if not faster, and we
don't have to clog the CPU's L2 cache with a tens-of-KB pointer array.
If we ever decide to adopt a more dynamic allocation method for shared
buffers, we'll probably have to revert this patch, but in the meantime
we might as well save a few bytes and nanoseconds. Per Qingqing Zhou.
exit, instead of trying to take shortcuts. Introduce some additional
shutdown callback routines to eliminate kluges like having ProcKill
be responsible for shutting down the buffer manager. Ensure that the
order of operations during shutdown is predictable and what you would
expect given the module layering.
syntactic conflicts, both privilege and role GRANT/REVOKE commands have
to use the same production for scanning the list of tokens that might
eventually turn out to be privileges or role names. So, change the
existing GRANT/REVOKE code to expect a list of strings not pre-reduced
AclMode values. Fix a couple other minor issues while at it, such as
InitializeAcl function name conflicting with a Windows system function.
includes error checking and an appropriate ereport(ERROR) message.
This gets rid of rather tedious and error-prone manipulation of errno,
as well as a Windows-specific bug workaround, at more than a dozen
call sites. After an idea in a recent patch by Heikki Linnakangas.
it is sufficient to track whether a backend holds a lock or not, and
store information about transaction vs. session locks only in the
inside-the-backend LocalLockTable. Since there can now be but one
PROCLOCK per lock per backend, LockCountMyLocks() is no longer needed,
thus eliminating some O(N^2) behavior when a backend holds many locks.
Also simplify the LockAcquire/LockRelease API by passing just a
'sessionLock' boolean instead of a transaction ID. The previous API
was designed with the idea that per-transaction lock holding would be
important for subtransactions, but now that we have subtransactions we
know that this is unwanted. While at it, add an 'isTempObject' parameter
to LockAcquire to indicate whether the lock is being taken on a temp
table. This is not used just yet, but will be needed shortly for
two-phase commit.
a descriptor that uses the current transaction snapshot, rather than
SnapshotNow as it did before (and still does if INV_WRITE is set).
This means pg_dump will now dump a consistent snapshot of large object
contents, as it never could do before. Also, add a lo_create() function
that is similar to lo_creat() but allows the desired OID of the large
object to be specified. This will simplify pg_restore considerably
(but I'll fix that in a separate commit).
are now reported via elog, eliminating the need to test the result code
at most call sites. Make it possible for the caller to distinguish a
freshly acquired lock from one already held in the current transaction.
Use that capability to avoid redundant AcceptInvalidationMessages() calls
in LockRelation().
Also, remove the rather useless return value of LockReleaseAll. Change
response to detection of corruption in the shared lock tables to PANIC,
since that is the only way of cleaning up fully.
Originally an idea of Heikki Linnakangas, variously hacked on by
Alvaro Herrera and Tom Lane.
communication structure, and make it its own module with its own lock.
This should reduce contention at least a little, and it definitely makes
the code seem cleaner. Per my recent proposal.
is contention for a tuple-level lock. This solves the problem of a
would-be exclusive locker being starved out by an indefinite succession
of share-lockers. Per recent discussion with Alvaro.
Essentially, we shoehorn in a lockable-object-type field by taking
a byte away from the lockmethodid, which can surely fit in one byte
instead of two. This allows less artificial definitions of all the
other fields of LOCKTAG; we can get rid of the special pg_xactlock
pseudo-relation, and also support locks on individual tuples and
general database objects (including shared objects). None of those
possibilities are actually exploited just yet, however.
I removed pg_xactlock from pg_class, but did not force initdb for
that change. At this point, relkind 's' (SPECIAL) is unused and
could be removed entirely.
to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE
paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU
data structure (managed much like pg_subtrans) to represent multiple-
transaction-ID sets. When more than one transaction is holding a shared
lock on a particular row, we create a MultiXactId representing that set
of transactions and store its ID in the row's XMAX. This scheme allows
an effectively unlimited number of row locks, just as we did before,
while not costing any extra overhead except when a shared lock actually
has to be shared. Still TODO: use the regular lock manager to control
the grant order when multiple backends are waiting for a row lock.
Alvaro Herrera and Tom Lane.
PageIndexTupleDelete() with a single pass of compactification ---
logic mostly lifted from PageRepairFragmentation. I noticed while
profiling that a VACUUM that's cleaning up a whole lot of deleted
tuples would spend as much as a third of its CPU time in
PageIndexTupleDelete; not too surprising considering the loop method
was roughly O(N^2) in the number of tuples involved.
to write out data that we are about to tell the filesystem to drop.
smgr_internal_unlink already had a DropRelFileNodeBuffers call to
get rid of dead buffers without a write after it's no longer possible
to roll back the deleting transaction. Adding a similar call in
smgrtruncate simplifies callers and makes the overall division of
labor clearer. This patch removes the former behavior that VACUUM
would write all dirty buffers of a relation unconditionally.
the freelist, plus per-buffer spinlocks that protect access to individual
shared buffer headers. This requires abandoning a global freelist (since
the freelist is a global contention point), which shoots down ARC and 2Q
as well as plain LRU management. Adopt a clock sweep algorithm instead.
Preliminary results show substantial improvement in multi-backend situations.
so that we can get the size of a shared inval message back down to what it
was in 7.4 (and simplify the logic too). Phase 2 of fixing the
'SMgrRelation hashtable corrupted' problem.
is the minimum required fix. I want to look next at taking advantage of
it by simplifying the message semantics in the shared inval message queue,
but that part can be held over for 8.1 if it turns out too ugly.
Also performed an initial run through of upgrading our Copyright date to
extend to 2005 ... first run here was very simple ... change everything
where: grep 1996-2004 && the word 'Copyright' ... scanned through the
generated list with 'less' first, and after, to make sure that I only
picked up the right entries ...
to shared memory as soon as possible, ie, right after read_backend_variables.
The effective difference from the original code is that this happens
before instead of after read_nondefault_variables(), which loads GUC
information and is apparently capable of expanding the backend's memory
allocation more than you'd think it should. This should fix the
failure-to-attach-to-shared-memory reports we've been seeing on Windows.
Also clean up a few bits of unnecessarily grotty EXEC_BACKEND code.
shared memory segment ID. If we can't access the existing shmem segment,
it must not be relevant to our data directory. If we can access it,
then attach to it and check for an actual match to the data directory.
This should avoid some cases of failure-to-restart-after-boot without
introducing any significant risk of failing to detect a still-running
old backend.
pins at end of transaction, and reduce AtEOXact_Buffers to an Assert
cross-check that this was done correctly. When not USE_ASSERT_CHECKING,
AtEOXact_Buffers is a complete no-op. This gets rid of an O(NBuffers)
bottleneck during transaction commit/abort, which recent testing has shown
becomes significant above a few tens of thousands of shared buffers.
(if any) currently waited for by LockBufferForCleanup(), which is all
that we were using it for anymore. Saves some space and eliminates
proportional-to-NBuffers slowdown in UnlockBuffers().
http://archives.postgresql.org/pgsql-hackers/2004-10/msg00464.php.
This fix is intended to be permanent: it moves the responsibility for
calling SetBufferCommitInfoNeedsSave() into the tqual.c routines,
eliminating the requirement for callers to test whether t_infomask changed.
Also, tighten validity checking on buffer IDs in bufmgr.c --- several
routines were paranoid about out-of-range shared buffer numbers but not
about out-of-range local ones, which seems a tad pointless.
as per recent discussions. Invent SubTransactionIds that are managed like
CommandIds (ie, counter is reset at start of each top transaction), and
use these instead of TransactionIds to keep track of subtransaction status
in those modules that need it. This means that a subtransaction does not
need an XID unless it actually inserts/modifies rows in the database.
Accordingly, don't assign it an XID nor take a lock on the XID until it
tries to do that. This saves a lot of overhead for subtransactions that
are only used for error recovery (eg plpgsql exceptions). Also, arrange
to release a subtransaction's XID lock as soon as the subtransaction
exits, in both the commit and abort cases. This avoids holding many
unique locks after a long series of subtransactions. The price is some
additional overhead in XactLockTableWait, but that seems acceptable.
Finally, restructure the state machine in xact.c to have a more orthogonal
set of states for subtransactions.
PROCLOCK structs in shared memory now have only a bitmask for held
locks, rather than counts (making them 40 bytes smaller, which is a
good thing). Multiple locks within a transaction are counted in the
local hash table instead, and we have provision for tracking which
ResourceOwner each count belongs to. Solves recently reported problem
with memory leakage within long transactions.
updates are no longer WAL-logged nor even fsync'd; we do not need to,
since after a crash no old pg_subtrans data is needed again. We truncate
pg_subtrans to RecentGlobalXmin at each checkpoint. slru.c's API is
refactored a little bit to separate out the necessary decisions.
for transaction commits that occurred just before the checkpoint. This is
an EXTREMELY serious bug --- kudos to Satoshi Okada for creating a
reproducible test case to prove its existence.
number of active subtransaction XIDs in each backend's PGPROC entry,
and use this to avoid expensive probes into pg_subtrans during
TransactionIdIsInProgress. Extend EOXactCallback API to allow add-on
modules to get control at subxact start/end. (This is deliberately
not compatible with the former API, since any uses of that API probably
need manual review anyway.) Add basic reference documentation for
SAVEPOINT and related commands. Minor other cleanups to check off some
of the open issues for subtransactions.
Alvaro Herrera and Tom Lane.
password/group files. Also allow read-only subtransactions of a read-write
parent, but not vice versa. These are the reasonably noncontroversial
parts of Alvaro's recent mop-up patch, plus further work on large objects
to minimize use of the TopTransactionResourceOwner.
recovery more manageable. Also, undo recent change to add FILE_HEADER
and WASTED_SPACE records to XLOG; instead make the XLOG page header
variable-size with extra fields in the first page of an XLOG file.
This should fix the boundary-case bugs observed by Mark Kirkwood.
initdb forced due to change of XLOG representation.
keep track of portal-related resources separately from transaction-related
resources. This allows cursors to work in a somewhat sane fashion with
nested transactions. For now, cursor behavior is non-subtransactional,
that is a cursor's state does not roll back if you abort a subtransaction
that fetched from the cursor. We might want to change that later.
performance front, but with feature freeze upon us I think it's time to
drive a stake in the ground and say that this will be in 7.5.
Alvaro Herrera, with some help from Tom Lane.
to reference the spinlock variable, and specify "memory" as a clobber
operand to be sure gcc does not try to keep shared-memory values in
registers across a spinlock acquisition. Also tighten the S/390 asm
sequence, which was apparently written with only minimal study of the
gcc asm documentation. I have personally tested i386, ia64, ppc, hppa,
and s390 variants --- there is some small chance that I broke the others,
but I doubt it.
There are various things left to do: contrib dbsize and oid2name modules
need work, and so does the documentation. Also someone should think about
COMMENT ON TABLESPACE and maybe RENAME TABLESPACE. Also initlocation is
dead, it just doesn't know it yet.
Gavin Sherry and Tom Lane.
not holding the buffer's cntx_lock or io_in_progress_lock. A recent
report from Litao Wu makes me wonder whether it is ever possible for
us to drop a buffer and forget to release its cntx_lock. The Assert
does not fire in the regression tests, but that proves little ...
locking conflict against concurrent CHECKPOINT that was discussed a few
weeks ago. Also, if not using WAL archiving (which is always true ATM
but won't be if PITR makes it into this release), there's no need to
WAL-log the index build process; it's sufficient to force-fsync the
completed index before commit. This seems to gain about a factor of 2
in my tests, which is consistent with writing half as much data. I did
not try it with WAL on a separate drive though --- probably the gain would
be a lot less in that scenario.
temp tables, and avoid WAL-logging truncations of temp tables. Do issue
fsync on truncated files (not sure this is necessary but it seems like
a good idea).
rather than an error code, and does elog(ERROR) not elog(WARNING)
when it detects a problem. All callers were simply elog(ERROR)'ing on
failure return anyway, and I find it hard to envision a caller that would
not, so we may as well simplify the callers and produce the more useful
error message directly.
explicitly fsync'ing every (non-temp) file we have written since the
last checkpoint. In the vast majority of cases, the burden of the
fsyncs should fall on the bgwriter process not on backends. (To this
end, we assume that an fsync issued by the bgwriter will force out
blocks written to the same file by other processes using other file
descriptors. Anyone have a problem with that?) This makes the world
safe for WIN32, which ain't even got sync(2), and really makes the world
safe for Unixen as well, because sync(2) never had the semantics we need:
it offers no way to wait for the requested I/O to finish.
Along the way, fix a bug I recently introduced in xlog recovery:
file truncation replay failed to clear bufmgr buffers for the dropped
blocks, which could result in 'PANIC: heap_delete_redo: no block'
later on in xlog replay.
than being random pieces of other files. Give bgwriter responsibility
for all checkpoint activity (other than a post-recovery checkpoint);
so this child process absorbs the functionality of the former transient
checkpoint and shutdown subprocesses. While at it, create an actual
include file for postmaster.c, which for some reason never had its own
file before.
about a third, make it work on non-Windows platforms again. (But perhaps
I broke the WIN32 code, since I have no way to test that.) Fold all the
paths that fork postmaster child processes to go through the single
routine SubPostmasterMain, which takes care of resurrecting the state that
would normally be inherited from the postmaster (including GUC variables).
Clean up some places where there's no particularly good reason for the
EXEC and non-EXEC cases to work differently. Take care of one or two
FIXMEs that remained in the code.
(SIGUSR1, which we have not been using recently) instead of piggybacking
on SIGUSR2-driven NOTIFY processing. This has several good results:
the processing needed to drain the sinval queue is a lot less than the
processing needed to answer a NOTIFY; there's less contention since we
don't have a bunch of backends all trying to acquire exclusive lock on
pg_listener; backends that are sitting inside a transaction block can
still drain the queue, whereas NOTIFY processing can't run if there's
an open transaction block. (This last is a fairly serious issue that
I don't think we ever recognized before --- with clients like JDBC that
tend to sit with open transaction blocks, the sinval queue draining
mechanism never really worked as intended, probably resulting in a lot
of useless cache-reset overhead.) This is the last of several proposed
changes in response to Philip Warner's recent report of sinval-induced
performance problems.
costing us lots more to maintain than it was worth. On shared tables
it was of exactly zero benefit because we couldn't trust it to be
up to date. On temp tables it sometimes saved an lseek, but not often
enough to be worth getting excited about. And the real problem was that
we forced an lseek on every relcache flush in order to update the field.
So all in all it seems best to lose the complexity.
of whether we have successfully read data into a buffer; this makes the
error behavior a bit more transparent (IMHO anyway), and also makes it
work correctly for local buffers which don't use Start/TerminateBufferIO.
Collapse three separate functions for writing a shared buffer into one.
This overlaps a bit with cleanups that Neil proposed awhile back, but
seems not to have committed yet.
of VACUUM cases so that VACUUM requests don't affect the ARC state at all,
avoid corner case where BufferSync would uselessly rewrite a buffer that
no longer contains the page that was to be flushed. Make some minor
other cleanups in and around the bufmgr as well, such as moving PinBuffer
and UnpinBuffer into bufmgr.c where they really belong.
* removed a few redundant defines
* get_user_name safe under win32
* rationalized pipe read EOF for win32 (UPDATED PATCH USED)
* changed all backend instances of sleep() to pg_usleep
- except for the SLEEP_ON_ASSERT in assert.c, as it would exceed a
32-bit long [Note to patcher: If a SLEEP_ON_ASSERT of 2000 seconds is
acceptable, please replace with pg_usleep(2000000000L)]
I added a comment to that part of the code:
/*
* It would be nice to use pg_usleep() here, but only does 2000 sec
* or 33 minutes, which seems too short.
*/
sleep(1000000);
Claudio Natoli
* Changes incorrect CYGWIN defines to __CYGWIN__
* Some localtime returns NULL checks (when unchecked cause SEGVs under
Win32
regression tests)
* Rationalized CreateSharedMemoryAndSemaphores and
AttachSharedMemoryAndSemaphores (Bruce, I finally remembered to do it);
requires attention.
Claudio Natoli
and FreeDir routines modeled on the existing AllocateFile/FreeFile.
Like the latter, these routines will avoid failing on EMFILE/ENFILE
conditions whenever possible, and will prevent leakage of directory
descriptors if an elog() occurs while one is open.
Also, reduce PANIC to ERROR in MoveOfflineLogs() --- this is not
critical code and there is no reason to force a DB restart on failure.
All per recent trouble report from Olivier Hubaut.
number of openable files and the number already opened. This eliminates
depending on sysconf(_SC_OPEN_MAX), and allows much saner behavior on
platforms where open-file slots are used up by semaphores.
for already empty buffers because their buffer tag was not cleard out
when the buffers have been invalidated before.
Also removed the misnamed BM_FREE bufhdr flag and replaced the checks,
which effectively ask if the buffer is unpinned, with checks against the
refcount field.
Jan
wit: Add a header record to each WAL segment file so that it can be reliably
identified. Avoid splitting WAL records across segment files (this is not
strictly necessary, but makes it simpler to incorporate the header records).
Make WAL entries for file creation, deletion, and truncation (as foreseen but
never implemented by Vadim). Also, add support for making XLOG_SEG_SIZE
configurable at compile time, similarly to BLCKSZ. Fix a couple bugs I
introduced in WAL replay during recent smgr API changes. initdb is forced
due to changes in pg_control contents.
the relcache, and so the notion of 'blind write' is gone. This should
improve efficiency in bgwriter and background checkpoint processes.
Internal restructuring in md.c to remove the not-very-useful array of
MdfdVec objects --- might as well just use pointers.
Also remove the long-dead 'persistent main memory' storage manager (mm.c),
since it seems quite unlikely to ever get resurrected.
done by the background writer between writing dirty blocks and
napping.
none (default) no action
sync bgwriter calls smgrsync() causing a sync(2)
A global sync() is only good on dedicated database servers, so
more flush methods should be added in the future.
Jan
instruction in the s_lock() wait loop, and use test before test-and-set
in TAS() macro to avoid unnecessary bus traffic. Patch from Manfred
Spraul, reworked a bit by Tom.
that were broken, try to make layout of s_lock.h entries consistent,
use HAVE_SPINLOCKS in preference to HAS_TEST_AND_SET everywhere outside
s_lock.h itself.
some concurrent changes Jan was making to the bufmgr. Here's an
updated version of the patch -- it should apply cleanly to CVS
HEAD and passes the regression tests.
This patch makes the following changes:
- remove the UnlockAndReleaseBuffer() and UnlockAndWriteBuffer()
macros, and replace uses of them with calls to the appropriate
functions.
- remove a bunch of #ifdef BMTRACE code: it is ugly & broken
(i.e. it doesn't compile)
- make BufferReplace() return a bool, not an int
- cleanup some logic in bufmgr.c; should be functionality
equivalent to the previous code, just cleaner now
- remove the BM_PRIVATE flag as it is unused
- improve a few comments, etc.
on 64-bit Solaris. Use a non-system-dependent datatype for UsedShmemSegID,
namely unsigned long (which we were already assuming could hold a shmem
key anyway, cf RecordSharedMemoryInLockFile).
method control structure, or a table of control structures.
. Use type LOCKMASK where an int is not a counter.
. Get rid of INVALID_TABLEID, use INVALID_LOCKMETHOD instead.
. Use INVALID_LOCKMETHOD instead of (LOCKMETHOD) NULL, because
LOCKMETHOD is not a pointer.
. Define and use macro LockMethodIsValid.
. Rename LOCKMETHOD to LOCKMETHODID.
. Remove global variable LongTermTableId in lmgr.c, because it is
never used.
. Make LockTableId static in lmgr.c, because it is used nowhere else.
Why not remove it and use DEFAULT_LOCKMETHOD?
. Rename the lock method control structure from LOCKMETHODTABLE to
LockMethodData. Introduce a pointer type named LockMethod.
. Remove elog(FATAL) after InitLockTable() call in
CreateSharedMemoryAndSemaphores(), because if something goes wrong,
there is elog(FATAL) in LockMethodTableInit(), and if this doesn't
help, an elog(ERROR) in InitLockTable() is promoted to FATAL.
. Make InitLockTable() void, because its only caller does not use its
return value any more.
. Rename variables in lock.c to avoid statements like
LockMethodTable[NumLockMethods] = lockMethodTable;
lockMethodTable = LockMethodTable[lockmethod];
. Change LOCKMETHODID type to uint16 to fit into struct LOCKTAG.
. Remove static variables BITS_OFF and BITS_ON from lock.c, because
I agree to this doubt:
* XXX is a fetch from a static array really faster than a shift?
. Define and use macros LOCKBIT_ON/OFF.
Manfred Koizar
This first part of the background writer does no syncing at all.
It's only purpose is to keep the LRU heads clean so that regular
backends seldom to never have to call write().
Jan
Remove the 'strategy map' code, which was a large amount of mechanism
that no longer had any use except reverse-mapping from procedure OID to
strategy number. Passing the strategy number to the index AM in the
first place is simpler and faster.
This is a preliminary step in planned support for cross-datatype index
operations. I'm committing it now since the ScanKeyEntryInitialize()
API change touches quite a lot of files, and I want to commit those
changes before the tree drifts under me.
pghackers. This fixes the problem recently reported by Markus KrÌutner
(hash bucket split corrupts the state of scans being done concurrently),
and I believe it also fixes all the known problems with deadlocks in
hash index operations. Hash indexes are still not really ready for prime
time (since they aren't WAL-logged), but this is a step forward.
index pages: when _bt_getbuf asks the FSM for a free index page, it is
possible (and, in some cases, even moderately likely) that the answer
will be the same page that _bt_split is trying to split. _bt_getbuf
already knew that the returned page might not be free, but it wasn't
prepared for the possibility that even trying to lock the page could
be problematic. Fix by doing a conditional rather than unconditional
grab of the page lock.
that OS=hpux is the same as CPU=hppa. First steps at doing this.
With these patches, we still work on hppa with either gcc or HP's cc.
We might work on hpux/itanium with gcc, but I can't test it. Definitely
will not work on hpux/itanium with non-gcc compiler, for lack of spinlock
code.
least-recently-used strategy from clog.c into slru.c. It doesn't
change any visible behaviour and passes all regression tests plus a
TruncateCLOG test done manually.
Apart from refactoring I made a little change to SlruRecentlyUsed,
formerly ClogRecentlyUsed: It now skips incrementing lru_counts, if
slotno is already the LRU slot, thus saving a few CPU cycles. To make
this work, lru_counts are initialised to 1 in SimpleLruInit.
SimpleLru will be used by pg_subtrans (part of the nested transactions
project), so the main purpose of this patch is to avoid future code
duplication.
Manfred Koizar
page when it's read in, per pghackers discussion around 17-Feb. Add a
GUC variable zero_damaged_pages that causes the response to be a WARNING
followed by zeroing the page, rather than the normal ERROR; this is per
Hiroshi's suggestion that there needs to be a way to get at the data
in the rest of the table.
(materialization into a tuple store) discussed on pgsql-hackers earlier.
I've updated the documentation and the regression tests.
Notes on the implementation:
- I needed to change the tuple store API slightly -- it assumes that it
won't be used to hold data across transaction boundaries, so the temp
files that it uses for on-disk storage are automatically reclaimed at
end-of-transaction. I added a flag to tuplestore_begin_heap() to control
this behavior. Is changing the tuple store API in this fashion OK?
- in order to store executor results in a tuple store, I added a new
CommandDest. This works well for the most part, with one exception: the
current DestFunction API doesn't provide enough information to allow the
Executor to store results into an arbitrary tuple store (where the
particular tuple store to use is chosen by the call site of
ExecutorRun). To workaround this, I've temporarily hacked up a solution
that works, but is not ideal: since the receiveTuple DestFunction is
passed the portal name, we can use that to lookup the Portal data
structure for the cursor and then use that to get at the tuple store the
Portal is using. This unnecessarily ties the Portal code with the
tupleReceiver code, but it works...
The proper fix for this is probably to change the DestFunction API --
Tom suggested passing the full QueryDesc to the receiveTuple function.
In that case, callers of ExecutorRun could "subclass" QueryDesc to add
any additional fields that their particular CommandDest needed to get
access to. This approach would work, but I'd like to think about it for
a little bit longer before deciding which route to go. In the mean time,
the code works fine, so I don't think a fix is urgent.
- (semi-related) I added a NO SCROLL keyword to DECLARE CURSOR, and
adjusted the behavior of SCROLL in accordance with the discussion on
-hackers.
- (unrelated) Cleaned up some SGML markup in sql.sgml, copy.sgml
Neil Conway
at database shutdown, and then load it again at database startup. This
preserves our hard-won knowledge of free space across restarts (given
an orderly shutdown, that is).
Adjustable threshold is gone in favor of keeping track of total requested
page storage and doling out proportional fractions to each relation
(with a minimum amount per relation, and some quantization of the results
to avoid thrashing with small changes in page counts). Provide special-
case code for indexes so as not to waste space storing useless page
free space counts. Restructure internal data storage to be a flat array
instead of list-of-chunks; this may cost a little more work in data
copying when reorganizing, but allows binary search to be used during
lookup_fsm_page_entry().
RelOid_pg_class, and transaction locks XactLockTableId. RelId is renamed
to objId.
- LockObject() and UnlockObject() functions created, and their use
sprinkled throughout the code to do descent locking for domains and
types. They accept lock modes AccessShare and AccessExclusive, as we
only really need a 'read' and 'write' lock at the moment. Most locking
cases are held until the end of the transaction.
This fixes the cases Tom mentioned earlier in regards to locking with
Domains. If the patch is good, I'll work on cleaning up issues with
other database objects that have this problem (most of them).
Rod Taylor
>
> ... he is now about to write an inlined version that can go into
> s_lock.h . I'll send the new patch later on...
OK, here it comes:
An inlined version of tas(), that works for both, powerpc and
powerpc64. The patch is against 7.3b5 and passes the test suite on
both architectures.
Reinhard Max
between signal handler and enable/disable code, avoid accumulation of
timing error due to trying to maintain remaining-time instead of
absolute-end-time, disable timeout before commit not after.
during the regression test. The problem has been reproduced on two machine
but both of these are the same type of hardware and software. I also tried
to recreate the problem on other machines, on older version of AIX but I
couldn't.
After looked through pgsql-hackers mailing list, I focused on spin lock
issue to solve the problem. The easiest and may not be the best solution
for the problem is to give up HAS_TEST_AND_SET. This actually works.
One another and better solution for the problem is to use _check_lock() and
_clear_lock() as spin lock. Important thing here is to define S_UNLOCK()
with _clear_lock(). This will solve the so called "Compiler bug" issue
someone wrote on the mailing list.
We have some other API such as cs(), compare_and_swap() and fetch_and_or()
to do test and set on AIX, but any of these didn't solve my problem. I
wrote tiny testing program to see if we have any bug of these API of AIX,
but I couldn't see any problem except for compare_and_swap(). It seems that
you can not use compare_and_swap() for the purpose, as it would not work as
spin lock on any SMP machines I tested. I don't know the reason why cs()
nor fetch_and_or()/fetch_and_and() will not work with PostgreSQL on p690.
These worked with my testing program on all machines I tested.
Tomoyuki Niijima
(overlaying low byte of page size) and add HEAP_HASOID bit to t_infomask,
per earlier discussion. Simplify scheme for overlaying fields in tuple
header (no need for cmax to live in more than one place). Don't try to
clear infomask status bits in tqual.c --- not safe to do it there. Don't
try to force output table of a SELECT INTO to have OIDs, either. Get rid
of unnecessarily complex three-state scheme for TupleDesc.tdhasoids, which
has already caused one recent failure. Improve documentation.
available (else there's no way to interpret the list links). Change
pg_locks view to show transaction ID locks separately from ordinary
relation locks. Avoid showing N duplicate rows when the same lock is
held multiple times (seems unlikely that users care about exact hold
count). Improve documentation.
connections by the superuser only.
This patch replaces the last patch I sent a couple of days ago.
It closes a connection that has not been authorised by a superuser if it would
leave less than the GUC variable ReservedBackends
(superuser_reserved_connections in postgres.conf) backend process slots free
in the SISeg. This differs to the first patch which only reserved the last
ReservedBackends slots in the procState array. This has made the free slot
test more expensive due to the use of a lock.
After thinking about a comment on the first patch I've also made it a fatal
error if the number of reserved slots is not less than the maximum number of
connections.
Nigel J. Andrews
This patch is an updated version of the lock listing patch. I've made
the following changes:
- write documentation
- wrap the SRF in a view called 'pg_locks': all user-level
access should be done through this view
- re-diff against latest CVS
One thing I chose not to do is adapt the SRF to use the anonymous
composite type code from Joe Conway. I'll probably do that eventually,
but I'm not really convinced it's a significantly cleaner way to
bootstrap SRF builtins than the method this patch uses (of course, it
has other uses...)
Neil Conway
The local buffer manager is no longer used for newly-created relations
(unless they are TEMP); a new non-TEMP relation goes through the shared
bufmgr and thus will participate normally in checkpoints. But TEMP relations
use the local buffer manager throughout their lifespan. Also, operations
in TEMP relations are not logged in WAL, thus improving performance.
Since it's no longer necessary to fsync relations as they move out of the
local buffers into shared buffers, quite a lot of smgr.c/md.c/fd.c code
is no longer needed and has been removed: there's no concept of a dirty
relation anymore in md.c/fd.c, and we never fsync anything but WAL.
Still TODO: improve local buffer management algorithms so that it would
be reasonable to increase NLocBuffer.
>usefulness.
>
>> [...] should I post a patch that puts pagesize directly into
>> PageHeaderData?
>
>If you're so inclined. Given that pd_opaque is hidden in those macros,
>there wouldn't be much of any gain in readability either, so I haven't
>worried about changing the declaration.
Thanks for the clarification. Here is the patch. Not much gain, but at
least it saves the next junior hacker from scratching his head ...
Manfred Koizar
Simon Riggs
Heikki Linnakangas
Tom Lane
Magnus Hagander
Bruce Momjian
Robert Haas
Alvaro Herrera
Peter Eisentraut
Neil Conway
Tatsuo Ishii
PostgreSQL Daemon
Jan Wieck