to ensure that we have released buffer refcounts and so forth, rather than
putting ad-hoc operations before (some of the calls to) proc_exit. Add
commentary to discourage future hackers from repeating that mistake.
included by everything that includes bufmgr.h --- it's supposed to be
internals, after all, not part of the API! This fixes the conflict
against FreeBSD headers reported by Rosenman, by making it unnecessary
for s_lock.h to be included by plperl.c.
IPC key assignment will now work correctly even when multiple postmasters
are using same logical port number (which is possible given -k switch).
There is only one shared-mem segment per postmaster now, not 3.
Rip out broken code for non-TAS case in bufmgr and xlog, substitute a
complete S_LOCK emulation using semaphores in spin.c. TAS and non-TAS
logic is now exactly the same.
When deadlock is detected, "Deadlock detected" is now the elog(ERROR)
message, rather than a NOTICE that comes out before an unhelpful ERROR.
that search loops only have to scan that far and not through all maxBackends
entries. This eliminates a performance penalty for setting maxBackends
much higher than the average number of active backends. Also, eliminate
no-longer-used 'backend tag' concept. Remove setting of environment
variables at backend start (except for CYR_RECODE), since none of them
are being examined by the backend any longer.
(WAL logging for this is not done yet, however.) Clean up a number of really
crufty things that are no longer needed now that DROP behaves nicely. Make
temp table mapper do the right things when drop or rename affecting a temp
table is rolled back. Also, remove "relation modified while in use" error
check, in favor of locking tables at first reference and holding that lock
throughout the statement.
as full as possible, seems better to use a tuple size around BLCKSZ/4
so that less space is wasted when a LO tuple is updated. Also, this
lets us use a logical page size that's an exact power of two, avoiding
partial-page writes when client is sending us stuff in power-of-2
buffer chunks.
kibitzing from Tom Lane. Large objects are now all stored in a single
system relation "pg_largeobject" --- no more xinv or xinx files, no more
relkind 'l'. This should offer substantial performance improvement for
large numbers of LOs, since there won't be directory bloat anymore.
It'll also fix problems like running out of locktable space when you
access thousands of LOs in one transaction.
Also clean up cruft in read/write routines. LOs with "holes" in them
(never-written byte ranges) now work just like Unix files with holes do:
a hole reads as zeroes but doesn't occupy storage space.
INITDB forced!
from bufmgr - it would be nice to have separate hash in smgr
for node <--> fd mappings, but for the moment it's easy to
add new hash to relcache.
Fixed small bug in xlog.c:ReadRecord.
* Makefile: Add more standard targets. Improve shell redirection in GNU
make detection.
* src/backend/access/transam/rmgr.c: Fix incorrect(?) C.
* src/backend/libpq/pqcomm.c (StreamConnection): Work around accept() bug.
* src/include/port/unixware.h: ...with help from here.
* src/backend/nodes/print.c (plannode_type): Remove some "break"s after
"return"s.
* src/backend/tcop/dest.c (DestToFunction): ditto.
* src/backend/nodes/readfuncs.c: Add proper prototypes.
* src/backend/utils/adt/numutils.c (pg_atoi): Cope specially with strtol()
setting EINVAL. This saves us from creating an extra set of regression test
output for the affected systems.
* src/include/storage/s_lock.h (tas): Correct prototype.
* src/interfaces/libpq/fe-connect.c (parseServiceInfo): Don't use variable
as dimension in array definition.
* src/makefiles/Makefile.unixware: Add support for GCC.
* src/template/unixware: same here
* src/test/regress/expected/abstime-solaris-1947.out: Adjust whitespace.
* src/test/regress/expected/horology-solaris-1947.out: Part of this file
was evidently missing.
* src/test/regress/pg_regress.sh: Fix shell. mkdir -p returns non-zero if
the directory exists.
* src/test/regress/resultmap: Add entries for Unixware.
> Regression tests opr_sanity and sanity_check are now failing.
Um, Bruce, I've said several times that I didn't think Perchine's large
object changes should be applied until someone had actually reviewed
them.
I tested it restoring my database with > 100000 BLOBS, and dumping it out.
But unfortunatly I can not restore it back due to problems in pg_dump.
--
Sincerely Yours,
Denis Perchine
> > For a while I though it might be because we are using an alpha TAS in
> > the spinlock rather than the old semaphore. I replaced our spinlock
> > with the standard one and it made no difference. We have been running
> > with our spinlock implementation for nearly 2 months on a production
> > database now without a hitch, so I think it is ok. Did I ever submit
> > any patches for the Alpha spinlock?
>
> Not that I recall. We did get some advice from some Alpha gurus at DEC
> who seemed to think the existing TAS code is OK. What was it that you
> felt needed to be improved?
The current code uses semaphores, which has the advantage that it works
well even on multi-processor machines, but the disadvantage that it is not
the fastest way possible. Writing a spinlock on Alpha for SMP machines is
very difficult, as you need to deal with memory barriers. A real mess. But
then one of the people at Compaq pointed out to us that there is a
ready-made routine on Alpha. We implemented it with the two patches below.
I ran tests with lots of parallel back-ends and got around a 10% speed
increase. I include the two patches. Perhaps some of the other people
running Tru64 can have a look at these as well.
Cheers,
Adriaan Joubert
> this is patch v 0.4 to support transactions with BLOBs.
> All BLOBs are in one table. You need to make initdb.
>
> --
> Sincerely Yours,
> Denis Perchine
after that dynamic loading isn't working and shared memory handling is
broken.
Attached with this message, there is a Zip file which contain :
* beos.diff = patch file generated with difforig
* beos = folder with beos support files which need to be moved in /
src/backend/port
* expected = foler with three file for message and precision
difference in regression test
* regression.diff = rule problem (need to kill the backend manualy)
* dynloader = dynloader files (they are also in the pacth files,
but there is so much modification that I have join full files)
Everything works except a problem in 'rules' Is there some problems
with rules in the current tree ? It used to works with last week tree.
Cyril VELTER
working on the VERY latest version of BeOS. I'm sure there will be
alot of comments, but then if there weren't I'd be disappointed!
Thanks for your continuing efforts to get this into your tree.
Haven't bothered with the new files as they haven't changed.
BTW Peter, the compiler is "broken" about the bool define and so on.
I'm filing a bug report to try and get it addressed. Hopefully then we
can tidy up the code a bit.
I await the replies with interest :)
David Reid
problems with some bits of it, but when all the patches are in it'll build
and we can fix it from there :) I've got a version that builds and runs and
that is the basis for these patches.
The first file has the new additional files that are required,
template/beos
backend/port/dynloader/beos.c
backend/port/dynloader/beos.h
include/port/beos.h
makefiles/Makefile.beos
The second is a tarball of diffs against a few files. I've added sys/ipc.h
to configure and config.h via configure.in and config.h.in and then started
adding the check as this file isn't needed on BeOS and having loads of
#ifdef BEOS isn't as obvious as #ifdef HAVE_SYS_IPC_H and isn't as
autconf'ish :)
Files touched are
include/c.h
configure.in
include/config.h.in
include/storage/ipc.h
include/utils/int8.h
Let me know how these go. I'll await a response before submitting any more.
Any problems just get in touch.
David Reid
as MaxHeapAttributeNumber. Increase MaxAttrSize to something more
reasonable (given what it's used for, namely checking char(n) declarations,
I didn't make it the full 1G that it could theoretically be --- 10Mb
seemed a more reasonable number). Improve calculation of MaxTupleSize.
duplicate keys by letting search go to the left rather than right when an
equal key is seen at an upper tree level. Fix poor choice of page split
point (leading to insertion failures) that was forced by chaining logic.
Don't store leftmost key in non-leaf pages, since it's not necessary.
Don't create root page until something is first stored in the index, so an
unused index is now 8K not 16K. (Doesn't seem to be as easy to get rid of
the metadata page, unfortunately.) Massive cleanup of unreadable code,
fix poor, obsolete, and just plain wrong documentation and comments.
See src/backend/access/nbtree/README for the gory details.
There's now only one transition value and transition function.
NULL handling in aggregates is a lot cleaner. Also, use Numeric
accumulators instead of integer accumulators for sum/avg on integer
datatypes --- this avoids overflow at the cost of being a little slower.
Implement VARIANCE() and STDDEV() aggregates in the standard backend.
Also, enable new LIKE selectivity estimators by default. Unrelated
change, but as long as I had to force initdb anyway...
PostgreSQL-7.0.2 run on Linux for the Intel-IA64 architecture. It also
fixes a bug in the configure scripts that caused configure to fail on
the fcntl(F_SETLK) test.
This fix triggered a bug in the fcntl(F_SETLK) code of the Linux
kernel when used on unix domain sockets resulting in postmaster to
segfault immediately after startup. There is a fix available and
included in the kernel that will be on SuSE Linux 7.0, but kernels <=
2.2.16 still have this bug.
Reinhard Max
for details). It doesn't really do that much yet, since there are no
short-term memory contexts in the executor, but the infrastructure is
in place and long-term contexts are handled reasonably. A few long-
standing bugs have been fixed, such as 'VACUUM; anything' in a single
query string crashing. Also, out-of-memory is now considered a
recoverable ERROR, not FATAL.
Eliminate a large amount of crufty, now-dead code in and around
memory management.
Fix problem with holding off SIGTRAP, SIGSEGV, etc in postmaster and
backend startup.
inputs have been converted to newstyle. This should go a long way towards
fixing our portability problems with platforms where char and short
parameters are passed differently from int-width parameters. Still
more to do for the Alpha port however.
it will close VFDs if necessary to surmount ENFILE or EMFILE failures.
Make use of this in md.c, xlog.c, and user.c routines that were
formerly vulnerable to these failures. In particular, this should
handle failures of mdblindwrt() that have been observed under heavy
load conditions. (By golly, every other process on the system may
crash after Postgres eats up all the kernel FDs, but Postgres will
keep going!)
That means you can now set your options in either or all of $PGDATA/configuration,
some postmaster option (--enable-fsync=off), or set a SET command. The list of
options is in backend/utils/misc/guc.c, documentation will be written post haste.
pg_options is gone, so is that pq_geqo config file. Also removed were backend -K,
-Q, and -T options (no longer applicable, although -d0 does the same as -Q).
Added to configure an --enable-syslog option.
changed all callers from TPRINTF to elog(DEBUG)
Hiroshi. ReleaseRelationBuffers now removes rel's buffers from pool,
instead of merely marking them nondirty. The old code would leave valid
buffers for a deleted relation, which didn't cause any known problems
but can't possibly be a good idea. There were several places which called
ReleaseRelationBuffers *and* FlushRelationBuffers, which is now
unnecessary; but there were others that did not. FlushRelationBuffers
no longer emits a warning notice if it finds dirty buffers to flush,
because with the current bufmgr behavior that's not an unexpected
condition. Also, FlushRelationBuffers will flush out all dirty buffers
for the relation regardless of block number. This ensures that
pg_upgrade's expectations are met about tuple on-row status bits being
up-to-date on disk. Lastly, tweak BufTableDelete() to clear the
buffer's tag so that no one can mistake it for being a still-valid
buffer for the page it once held. Formerly, the buffer would not be
found by buffer hashtable searches after BufTableDelete(), but it would
still be thought to belong to its old relation by the routines that
sequentially scan the shared-buffer array. Again I know of no bugs
caused by that, but it still can't be a good idea.
whether to do fsync or not, and if so (which should be seldom) just
do the fsync immediately. This way we need not build data structures
in md.c/fd.c for blind writes.
as a shared dirtybit for each shared buffer. The shared dirtybit still
controls writing the buffer, but the local bit controls whether we need
to fsync the buffer's file. This arrangement fixes a bug that allowed
some required fsyncs to be missed, and should improve performance as well.
For more info see my post of same date on pghackers.
In the event of an elog() while the mode was set to immediate write,
there was no way for it to be set back to the normal delayed write.
The mechanism was a waste of space and cycles anyway, since the only user
was varsup.c, which could perfectly well call FlushBuffer directly.
Now it does just that, and the notion of a write mode is gone.
running gcc and HP's cc with warnings cranked way up. Signed vs unsigned
comparisons, routines declared static and then defined not-static,
that kind of thing. Tedious, but perhaps useful...
if presented an uninitialized (all zeroes) page. The system no longer
crashes hard if an all-zeroes page is present in a relation. There seem
to be some boundary conditions where a page will be appended to a relation
and zeroed, but its page header is never initialized; until we can track
down and fix all of those, robustness seems like a good idea.
Also, clean up some obsolete and downright wrong comments.
for the case of errors in backend startup, and proc_exit's method for
coping with errors during proc_exit was *completely* busted. Fixed per
discussions on pghackers around 11/6/99.
recycle storage within sort temp file on a block-by-block basis. This
reduces peak disk usage to essentially just the volume of data being
sorted, whereas it had been about 4x the data volume before.
BufFile so that it handles multi-segment temporary files transparently.
This allows sorts and hashes to work with data exceeding 2Gig (or whatever
the local limit on file size is). Change psort.c to use relative seeks
instead of absolute seeks for backwards scanning, so that it won't fail
when the data volume exceeds 2Gig.
* Buffer refcount cleanup (per my "progress report" to pghackers, 9/22).
* Add links to backend PROC structs to sinval's array of per-backend info,
and use these links for routines that need to check the state of all
backends (rather than the slow, complicated search of the ShmemIndex
hashtable that was used before). Add databaseOID to PROC structs.
* Use this to implement an interlock that prevents DESTROY DATABASE of
a database containing running backends. (It's a little tricky to prevent
a concurrently-starting backend from getting in there, since the new
backend is not able to lock anything at the time it tries to look up
its database in pg_database. My solution is to recheck that the DB is
OK at the end of InitPostgres. It may not be a 100% solution, but it's
a lot better than no interlock at all...)
* In ALTER TABLE RENAME, flush buffers for the relation before doing the
rename of the physical files, to ensure we don't get failures later from
mdblindwrt().
* Update TRUNCATE patch so that it actually compiles against current
sources :-(.
You should do "make clean all" after pulling these changes.
additional argument specifying the kind of lock to acquire/release (or
'NoLock' to do no lock processing). Ensure that all relations are locked
with some appropriate lock level before being examined --- this ensures
that relevant shared-inval messages have been processed and should prevent
problems caused by concurrent VACUUM. Fix several bugs having to do with
mismatched increment/decrement of relation ref count and mismatched
heap_open/close (which amounts to the same thing). A bogus ref count on
a relation doesn't matter much *unless* a SI Inval message happens to
arrive at the wrong time, which is probably why we got away with this
sloppiness for so long. Repair missing grab of AccessExclusiveLock in
DROP TABLE, ALTER/RENAME TABLE, etc, as noted by Hiroshi.
Recommend 'make clean all' after pulling this update; I modified the
Relation struct layout slightly.
Will post further discussion to pghackers list shortly.
offended my aesthestic sensibility that there was so much unreadable code
doing so little. Rewritten code is about half the size, faster, and
(I hope) much more intelligible.
the SInval spinlock while it is calling the passed invalFunction or
resetFunction. This is necessary to avoid deadlock with lmgr change;
InvalidateSharedInvalid can be called recursively now. It should be
a good performance improvement anyway --- holding a spinlock for more
than a very short interval is a no-no.
{
Oid relId;
Oid dbId;
union
{
BlockNumber blkno;
TransactionId xid;
} objId;
>
> Added:
> /*
> * offnum should be part of objId.tupleId above, but would increase
> * sizeof(LOCKTAG) and so moved here; currently used by userlocks only.
> */
> OffsetNumber offnum;
uint16 lockmethod; /* needed by userlocks */
} LOCKTAG;
gmake clean required...
User locks are ready for 6.5 release...
through MAXBACKENDS array entries used to be fine when MAXBACKENDS = 64.
It's not so cool with MAXBACKENDS = 1024 (or more!), especially not in a
frequently-used routine like SIDelExpiredDataEntries. Repair by making
procState array size be the soft MaxBackends limit rather than the hard
limit, and by converting SIGetProcStateLimit() to a macro.
files to be closed automatically at transaction abort or commit, should
they still be open. Also close any still-open stdio files allocated with
AllocateFile at abort/commit. This should eliminate problems with leakage
of file descriptors after an error. Also, put in some primitive buffered-IO
support so that psort.c can use virtual files without severe performance
penalties.
of MAXBACKENDS is now 1024, since all it's costing is about 32 bytes of memory
per array slot. configure's --with-maxbackends switch now controls DEF_MAXBACKENDS
which is simply the default value of the postmaster's -N switch. Thus,
the out-of-the-box configuration will still limit you to 64 backends,
but you can go up to 1024 backends simply by restarting the postmaster with
a different -N switch --- no rebuild required.
(--with-maxbackends). Add a postmaster switch (-N backends) that allows
the limit to be reduced at postmaster start time. (You can't increase it,
sorry to say, because there are still some fixed-size arrays.)
Grab the number of semaphores indicated by min(MAXBACKENDS, -N) at
postmaster startup, so that this particular form of bogus configuration
is exposed immediately rather than under heavy load.
spin-locks. Notice that it's now inline assembler in s_lock.h,
rather than seperate code in s_lock.c. It also shrank a little
bit... Just rip out the S_LOCK() define and insert the tas() inline
function. Please let me know if there are any problems with it.
Jon Buller
1. The UnixWare tas macro was reformatted (by indent or it like?) which caused
it to break. The asm macro construct is very particular about the %mem
construct -- it has to start in column 1.
2. When compiling libpq++, g++ was used even if configure found the C++ com-
piler to be CC.
3. When compiling libpq++, '-Wno-error' was added to CXXFLAGS, even if the
compiler wasn't g++.
Billy G. Allie
> sinval.patch
>
> fixes a problem in SI cache which causes table overflow if some
> backend is idle for a long time while other backends keep adding
> entries.
> It uses the new signal handling implemented in tprintf.patch.
> I have also increacasesed the max number of backends from 32 to 64
> and the table size from 1000 to 5000.
> I don't know if anybody is working on SI, but until another
> solution is found this patch fixes the problem. I have received
> messages from other people reporting the same problem which I
> fixed many months ago.
lock.patch
I have rewritten lock.c cleaning up the code and adding better
assert checking I have also added some fields to the lock and
xid tags for better support of user locks. There is also a new
function which returns an array of pids owning a lock.
I'm using this code from over six months and it works fine.
Summary of changes:
In pqcomm.h, use the SUN_LEN macro if it is defined to calculate
the size of the sockaddr_un structure.
In unixware.h, drop the use of the UNIXWARE macro. Everything can
be handled with the USE_UNIVEL_CC and DISABLE_COMPLEX_MACRO macros.
In s_lock.h, remove the reference to the UNIXWARE macro (see above).
In the unixware template, add the YFLAGS:-d line.
In various makefile templates, add (or cleanup) unixware and univel
port specific information.
-- Billy G. Allie
no longer returns buffer pointer, can be gotten from scan;
descriptor; bootstrap can create multi-key indexes;
pg_procname index now is multi-key index; oidint2, oidint4, oidname
are gone (must be removed from regression tests); use System Cache
rather than sequential scan in many places; heap_modifytuple no
longer takes buffer parameter; remove unused buffer parameter in
a few other functions; oid8 is not index-able; remove some use of
single-character variable names; cleanup Buffer variables usage
and scan descriptor looping; cleaned up allocation and freeing of
tuples; 18k lines of diff;
calls. Outside a transaction, the backend detects them as buffer
leaks; it sends a NOTICE, and frees them. This sometimes cause a
segmentation fault (at least on Linux). These indexes are initialized
on the first lo_read/lo_write/lo_tell call, and (normally) closed
on a lo_close call. Thus the buffer leaks appear when lo direct
access functions are used, and not with lo_import/lo_export functions
(libpq version calls lo_close before ending the command, and the
backend version uses another path).
The included patches (against recent snapshot, and against 6.3.2)
cause indexes to be closed on transaction end (that is on explicit
'END' statment, or on command termination outside trasaction blocks),
thus preventing the buffer leaks while increasing performance inside
transactions. Some (all?) 'classic' memory leaks are also removed.
I hope it will be ok.
--- Pascal ANDRE, graduated from Ecole Centrale Paris andre@via.ecp.fr
in a more readable form. -- I am submitting the following patches
to the June 6, 1998 snapshot of PostgreSQL. These patches implement
a port of PostgreSQL to SCO UnixWare 7, and updates the Univel port
(UnixWare 2.x). The patched files, and the reason
for the patch are:
File Reason for the patch ---------------
---------------------------------------------------------------
src/backend/port/dynloader/unixware.c src/backend/port/dynloader/unixware.h
src/include/port/unixware.h src/makefiles/Makefile.unixware
src/template/unixware
Created for the UNIXWARE port.
src/include/port/univel.h
Modifed this file to work with the changes made to
s_lock.[ch].
src/backend/storage/buffer/s_lock.c src/include/storage/s_lock.h
Moved the UNIXWARE (and Univel) tas() function from
s_lock.c to s_lock.h. The UnixWare compiler asm
construct is treated as a macro and needs to be in
the s_lock.h file. I also reworked the tas()
function to correct some errors in the code.
src/include/version.h.in
The use of the ## operator with quoted strings in
the VERSION macro caused problems with the UnixWare
C compiler. I removed the ## operators since they
were not needed in this case. The macro expands
into a sequence of quoted strings that will be
concatenated by any ANSI C compiler.
src/config.guess
This script was modified to recognize SCO UnixWare
7.
src/configure src/configure.in
The configure script was modified to recognize SCO
UnixWare 7.
Billy G. Allie
NS32K machine I contributed. In any case, I now have postgresql-6.3
running again on NetBSD/pc532, a NS32532 machine. The following
changes are needed relative to the src directory. (It looks like
support was partially removed when the files were moved from the
src/backend/storage/.... tree to the src/include tree.)
If you need me to get a current development version of postgresql
for this change let me know. Also, let me know if this code needs
updating due to another code movement that deleted the old NS32K
support.
Thank you.
Phil Nelson
GCC, the inner "#if defined(__GNUC__)" can just be omitted in that
architecture's block.
The existing arrangement with an outer "#if defined(__GNUC__)"
doesn't have any obvious benefit, and it encourages missed cases
like this one.
BTW, I'd suggest making the definition of clear_lock for HPUX be
static const slock_t clear_lock = {{-1, -1, -1, -1}};
The extra braces are needed to suppress warnings from gcc, and
declaring it const just seems like good practice.
regards, tom lane
but as I don't have access to a sparc for testing I just did what
I could. I am guessing here, but please apply the following to your
pgsql and let me know what happens. Also, cd to src/storage/buffer
and do 'make s_lock_test' as well.
David Gould
This incorporates all the precedeing patches and emailed suggestions
and the results of the performance testing I posted last week. I
would like to get this tested on as many platforms as possible so
I can verify it went in correctly (as opposed to the horrorshow
last time I sent in a patch).
Once this is confirmed, I will make a tarball of files that can be
dropped into a 6.3.2 source tree as a few people have asked for
this in 6.3.2 as well.
David Gould
Attached patch will add a version() function to Postges, e.g.
template1=> select version();
version
------------------------------------------------------------
PostgreSQL 6.3.2 on i586-pc-linux-gnu, compiled by gcc 2.8.1
(1 row)
Ok, I have finally gotten all of the defines for Dec/Alpha and
Linux/Alpha sorted out as Marc asked. There is no longer any need for
'-Dalpha' or '-Dlinuxalpha' in either the Dec/Alpha or the Linux/Alpha
template files (./src/template/{alpha,linuxalpha}). I have replaced every
instance of 'alpha' or '__alpha__' with '__alpha', as that appears to be
the common symbol between C compilers on both operating systems (RH4.2 &
DecUnix 4.0b) for alpha.
