1. Backends waiting for buffer I/O are now interruptible.
2. If something goes wrong in a backend that is currently performing
I/O, waiting backends no longer wake up until that backend reaches
AbortBufferIO() and broadcasts on the CV. Previously, any waiters would
wake up (because the I/O lock was automatically released) and then
busy-loop until AbortBufferIO() cleared BM_IO_IN_PROGRESS.
3. LWLockMinimallyPadded is removed, as it would now be unused.
Author: Robert Haas <robertmhaas@gmail.com>
Reviewed-by: Thomas Munro <thomas.munro@gmail.com>
Reviewed-by: Julien Rouhaud <rjuju123@gmail.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us> (earlier version, 2016)
Discussion: https://postgr.es/m/CA%2BhUKGJ8nBFrjLuCTuqKN0pd2PQOwj9b_jnsiGFFMDvUxahj_A%40mail.gmail.com
Discussion: https://postgr.es/m/CA+Tgmoaj2aPti0yho7FeEf2qt-JgQPRWb0gci_o1Hfr=C56Xng@mail.gmail.com
When a foreign key constraint is applied to a partitioned table, each
leaf partition inherits a similar FK constraint. We were processing all
of those constraints independently, meaning that in large partitioning
trees we'd build up large collections of cached FK-checking query plans.
However, in all cases but one, the generated queries are actually
identical for all members of the inheritance tree (because, in most
cases, the query only mentions the topmost table of the other side of
the FK relationship). So we can share a single cached plan among all
the partitions, saving memory, not to mention time to build and maintain
the cached plans.
Keisuke Kuroda and Amit Langote
Discussion: https://postgr.es/m/cab4b85d-9292-967d-adf2-be0d803c3e23@nttcom.co.jp_1
In the current lazy vacuum implementation, some index AMs such as
btree indexes call lazy_tid_reaped() for each index tuple during
ambulkdelete to check if the index tuple points to the (collected)
garbage tuple. In that function, we simply call bsearch(), but we
should be able to know the result without bsearch() if the index tuple
points to the heap tuple that is out of range of the collected garbage
tuples. Therefore, add a simple bound check before resorting to
bsearch(). Testing has shown that this can give significant
performance benefits.
Author: Masahiko Sawada <masahiko.sawada@2ndquadrant.com>
Discussion: https://www.postgresql.org/message-id/flat/CA+fd4k76j8jKzJzcx8UqEugvayaMSnQz0iLUt_XgBp-_-bd22A@mail.gmail.com
Commit 547f04e7 produced errors on AIX/xlc while building plpython. The
new code appears to be incompatible with the hack installed by commit
a11cf433. Without access to an AIX system to check, my guess is that
_POSIX_C_SOURCE may be required for <time.h> to declare the things the
header needs to see, but plpython.h undefines it.
For now, to unbreak build farm animal hoverfly, just move the new
pg_time_usec_t support into pgbench.c. Perhaps later we could figure
out what to rearrange to put it back into a header for wider use.
Discussion: https://postgr.es/m/CA%2BhUKG%2BP%2BjcD%3Dx9%2BagyTdWtjpOT64MYiGic%2Bcbu_TD8CV%3D6A3w%40mail.gmail.com
1. pg_time_usec_t needs to be printed with INT64_FORMAT, not %ld, or 32
bit systems complain, per lapwing.
2. Some Windows compilers didn't like a thread function not marked with
__stdcall, per whelk; let's see if this fixes the problem.
Moving this logic into pg_regress fixes a potential failure with
parallel tests when pg_upgrade and the main regression test suite both
trigger the makefile rule that cleaned up testtablespace/ under
src/test/regress. Even if pg_upgrade was triggering this rule, it has
no need to do so as it uses a different tablespace path. So if
pg_upgrade triggered the makefile rule for the tablespace setup while
the main regression test suite ran the tablespace cases, it would fail.
61be85a was a similar attempt at achieving that, but that broke cases
where the regression tests require to run under an Administrator
account, like with Appveyor.
Reported-by: Andres Freund, Kyotaro Horiguchi
Reviewed-by: Peter Eisentraut
Discussion: https://postgr.es/m/20201209012911.uk4d6nxcnkp7ehrx@alap3.anarazel.de
Wait until all pgbench threads are connected before benchmarking begins.
This fixes a problem where some connections could take a very long time
to be established because of lock contention from earlier connections,
making results unstable and bogus with high connection counts.
Author: Andres Freund <andres@anarazel.de>
Author: Fabien COELHO <coelho@cri.ensmp.fr>
Reviewed-by: Marina Polyakova <m.polyakova@postgrespro.ru>
Reviewed-by: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Reviewed-by: Hayato Kuroda <kuroda.hayato@fujitsu.com>
Reviewed-by: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200227180100.zyvjwzcpiokfsqm2%40alap3.anarazel.de
Instead of instr_time (struct timespec) and the INSTR_XXX macros,
introduce pg_time_usec_t and use integer arithmetic. Don't include the
connection time in TPS unless using -C mode, but report it separately.
Author: Fabien COELHO <coelho@cri.ensmp.fr>
Reviewed-by: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Reviewed-by: Hayato Kuroda <kuroda.hayato@fujitsu.com>
Discussion: https://postgr.es/m/20200227180100.zyvjwzcpiokfsqm2%40alap3.anarazel.de
Instead of maintaining an incomplete emulation of POSIX threads for
Windows, let's use an extremely minimalist macro-based abstraction for
now. A later patch will extend this, without the need to supply more
complicated pthread emulation code. (There may be a need for a more
serious portable thread abstraction in later projects, but this is not
it.)
Minor incidental problems fixed: it wasn't OK to use (pthread_t) 0 as a
special value, it wasn't OK to compare thread_t values with ==, and we
incorrectly assumed that pthread functions set errno.
Discussion: https://postgr.es/m/20200227180100.zyvjwzcpiokfsqm2%40alap3.anarazel.de
Initialize other newly added variables in max_parallel_hazard_context via
is_parallel_safe() because we don't check the parallel-safety of target
relations in that function.
Reported-by: Tom Lane as per buildfarm
Author: Amit Kapila
Discussion: https://postgr.es/m/2060179.1615347455@sss.pgh.pa.us
Parallel SELECT can't be utilized for INSERT in the following cases:
- INSERT statement uses the ON CONFLICT DO UPDATE clause
- Target table has a parallel-unsafe: trigger, index expression or
predicate, column default expression or check constraint
- Target table has a parallel-unsafe domain constraint on any column
- Target table is a partitioned table with a parallel-unsafe partition key
expression or support function
The planner is updated to perform additional parallel-safety checks for
the cases listed above, for determining whether it is safe to run INSERT
in parallel-mode with an underlying parallel SELECT. The planner will
consider using parallel SELECT for "INSERT INTO ... SELECT ...", provided
nothing unsafe is found from the additional parallel-safety checks, or
from the existing parallel-safety checks for SELECT.
While checking parallel-safety, we need to check it for all the partitions
on the table which can be costly especially when we decide not to use a
parallel plan. So, in a separate patch, we will introduce a GUC and or a
reloption to enable/disable parallelism for Insert statements.
Prior to entering parallel-mode for the execution of INSERT with parallel
SELECT, a TransactionId is acquired and assigned to the current
transaction state. This is necessary to prevent the INSERT from attempting
to assign the TransactionId whilst in parallel-mode, which is not allowed.
This approach has a disadvantage in that if the underlying SELECT does not
return any rows, then the TransactionId is not used, however that
shouldn't happen in practice in many cases.
Author: Greg Nancarrow, Amit Langote, Amit Kapila
Reviewed-by: Amit Langote, Hou Zhijie, Takayuki Tsunakawa, Antonin Houska, Bharath Rupireddy, Dilip Kumar, Vignesh C, Zhihong Yu, Amit Kapila
Tested-by: Tang, Haiying
Discussion: https://postgr.es/m/CAJcOf-cXnB5cnMKqWEp2E2z7Mvcd04iLVmV=qpFJrR3AcrTS3g@mail.gmail.com
Discussion: https://postgr.es/m/CAJcOf-fAdj=nDKMsRhQzndm-O13NY4dL6xGcEvdX5Xvbbi0V7g@mail.gmail.com
This partially reverts 096bbf7 and 9d2d457, undoing the libpq changes as
it could cause breakages in distributions that share one single libpq
version across multiple major versions of Postgres for extensions and
applications linking to that.
Note that the backend is unchanged here, and it still disables SSL
compression while simplifying the underlying catalogs that tracked if
compression was enabled or not for a SSL connection.
Per discussion with Tom Lane and Daniel Gustafsson.
Discussion: https://postgr.es/m/YEbq15JKJwIX+S6m@paquier.xyz
The authtype parameter was deprecated and made inactive in commit
d5bbe2aca5, but the environment variable was left defined and thus
tested with a getenv call even though the value is of no use. Also,
if it would exist it would be copied but never freed as the cleanup
code had been removed.
tty was deprecated in commit cb7fb3ca95 but most of the
infrastructure around it remained in place.
Author: Daniel Gustafsson <daniel@yesql.se>
Discussion: https://postgr.es/m/DDDF36F3-582A-4C02-8598-9B464CC42B34@yesql.se
Per buildfarm member crake, any servers including a postgres_fdw server
with this option set would fail to do a pg_upgrade properly as the
option got hidden in f9264d1 by becoming a debug option, making the
restore of the FDW server fail.
This changes back the option in libpq to be visible, but still inactive
to fix this upgrade issue.
Discussion: https://postgr.es/m/YEbq15JKJwIX+S6m@paquier.xyz
This commit adds new GUC track_wal_io_timing. When this is enabled,
the total amounts of time XLogWrite writes and issue_xlog_fsync syncs
WAL data to disk are counted in pg_stat_wal. This information would be
useful to check how much WAL write and sync affect the performance.
Enabling track_wal_io_timing will make the server query the operating
system for the current time every time WAL is written or synced,
which may cause significant overhead on some platforms. To avoid such
additional overhead in the server with track_io_timing enabled,
this commit introduces track_wal_io_timing as a separate parameter from
track_io_timing.
Note that WAL write and sync activity by walreceiver has not been tracked yet.
This commit makes the server also track the numbers of times XLogWrite
writes and issue_xlog_fsync syncs WAL data to disk, in pg_stat_wal,
regardless of the setting of track_wal_io_timing. This counters can be
used to calculate the WAL write and sync time per request, for example.
Bump PGSTAT_FILE_FORMAT_ID.
Bump catalog version.
Author: Masahiro Ikeda
Reviewed-By: Japin Li, Hayato Kuroda, Masahiko Sawada, David Johnston, Fujii Masao
Discussion: https://postgr.es/m/0509ad67b585a5b86a83d445dfa75392@oss.nttdata.com
The command (TO or FROM), its type (file, pipe, program or callback),
and the number of tuples excluded by a WHERE clause in COPY FROM are
added to the progress reporting already available.
The column "lines_processed" is renamed to "tuples_processed" to
disambiguate the meaning of this column in the cases of CSV and BINARY
COPY and to be more consistent with the other catalog progress views.
Bump catalog version, again.
Author: Matthias van de Meent
Reviewed-by: Michael Paquier, Justin Pryzby, Bharath Rupireddy, Josef
Šimánek, Tomas Vondra
Discussion: https://postgr.es/m/CAEze2WiOcgdH4aQA8NtZq-4dgvnJzp8PohdeKchPkhMY-jWZXA@mail.gmail.com
PostgreSQL disabled compression as of e3bdb2d and the documentation
recommends against using it since. Additionally, SSL compression has
been disabled in OpenSSL since version 1.1.0, and was disabled in many
distributions long before that. The most recent TLS version, TLSv1.3,
disallows compression at the protocol level.
This commit removes the feature itself, removing support for the libpq
parameter sslcompression (parameter still listed for compatibility
reasons with existing connection strings, just ignored), and removes
the equivalent field in pg_stat_ssl and de facto PgBackendSSLStatus.
Note that, on top of removing the ability to activate compression by
configuration, compression is actively disabled in both frontend and
backend to avoid overrides from local configurations.
A TAP test is added for deprecated SSL parameters to check after
backwards compatibility.
Bump catalog version.
Author: Daniel Gustafsson
Reviewed-by: Peter Eisentraut, Magnus Hagander, Michael Paquier
Discussion: https://postgr.es/m/7E384D48-11C5-441B-9EC3-F7DB1F8518F6@yesql.se
Throw a "function protocol violation" error if a function in FROM
tries to return a set though it wasn't marked proretset. Although
such cases work at the moment, it doesn't seem like something we
want to guarantee will keep working. Besides, there are other
negative consequences of not setting the proretset flag, such as
potentially bad plans.
No back-patch, since if there is any third-party code violating
this expectation, people wouldn't appreciate us breaking it in
a minor release.
Discussion: https://postgr.es/m/1636062.1615141782@sss.pgh.pa.us
The initial catalog data for this function failed to set proretset
or provide a prorows estimate. It accidentally worked anyway when
invoked in the FROM clause, because the executor isn't too picky
about this; but the planner didn't expect the function to return
multiple rows, which could lead to bad plans. Also the function
would fail if invoked in the SELECT list.
We can't easily back-patch this fix, but fortunately the bug's
consequences aren't awful in most cases. Getting this right is
mainly an exercise in future-proofing.
Discussion: https://postgr.es/m/1636062.1615141782@sss.pgh.pa.us
"SELECT pg_import_system_collations(0)" caused an assertion failure.
With a random nonzero argument --- or indeed with zero, in non-assert
builds --- it would happily make pg_collation entries with garbage
values of collnamespace. These are harmless as far as I can tell
(unless maybe the OID happens to become used for a schema, later on?).
In any case this isn't a security issue, since the function is
superuser-only. But it seems like a gotcha for unwary DBAs, so let's
add a check that the given OID belongs to some schema.
Back-patch to v10 where this function was introduced.
Coverity is still unhappy after commit 190c79884, and after looking
closer I think it might be onto something. The callers of newdfa()
typically drop out if v->err has been set nonzero, which newdfa()
is faithfully doing if it fails. However, what if v->err was already
nonzero before we entered newdfa()? Then newdfa() could succeed and
the caller would promptly leak its result.
I don't think this scenario can actually happen, but the predicate
"v->err is always zero when newdfa() is called" seems difficult to be
entirely sure of; there's a good deal of code that potentially could
get that wrong.
It seems better to adjust the callers to directly check for a null
result instead of relying on ISERR() tests. This is slightly cheaper
than the previous coding anyway.
Lacking evidence that there's any real bug, no back-patch.
Commit 1eb6d6527a allowed to track replica origin replay progress for 2PC
but it was not complete. It misses to properly track the progress for
rollback prepared especially it missed updating the code for recovery.
Additionally, we need to allow tracking it on subscriber nodes where
wal_level might not be logical.
It is required to track decoding of 2PC which is committed in PG14
(a271a1b50e) and also nobody complained about this till now so not
backpatching it.
Author: Amit Kapila
Reviewed-by: Michael Paquier and Ajin Cherian
Discussion: https://postgr.es/m/CAA4eK1L-kHmMnSdrRW6UhRbCjR7cgh04c+6psY15qzT6ktcd+g@mail.gmail.com
Using pgbench in an environment with both PGPORT and PGUSER set would
have caused the generation of a debug log with an incorrect database
name due to an oversight in 412893b. Not specifying user, port and/or
database using the option switches, without their respective environment
variables, generated a log entry with empty strings, which was
rather useless.
This commit fixes this set of issues by simplifying the logic grabbing
the connection information, removing a set of getenv() calls that
emulated what libpq already does. The faulty debug log now directly
uses the information from the libpq connection, and it gets generated
after the connection to the backend is completed, not before it (in the
event of a failure libpq would complain with more information about the
connection attempt so the log is not really useful before anyway).
Author: Kota Miyake
Reviewed-by: Fujii Masao, Michael Paquier
Discussion: https://postgr.es/m/026b3ae6fc339a18394d053c32a4463d@oss.nttdata.com
These can be set in buildenv.pl or a "set" command within a Windows
terminal. The MSVC script vcregress.pl parses the values available in
the environment to build the resulting prove commands, and the parsing
of PROVE_TESTS is able to handle name patterns in the same way as other
platforms.
Not specifying those environment values makes vcregress.pl fall back to
the previous default, with no extra flags for the prove command, and all
the tests run within t/.
Author: Michael Paquier
Reviewed-by: Juan José Santamaría Flecha, Julien Rouhaud
Discussion: https://postgr.es/m/YD9GigwHoL6lFY2y@paquier.xyz
Under windows, psql processes need to be ended explicitly, or the TAP
tests hang. However, the recovery tests were doing this via
IPC::Run::kill_kill(), which causes other major problems on Windows.
We solve this by instead sending '\q' to psql so it quits of its own
accord, and then simply waiting for it. This means we can now run almost
all the recovery tests on all Windows platforms.
Discussion: https://postgr.es/m/20210301200715.tdjpuesfzebpffgn@alap3.anarazel.de
Mistake in f06b1c598254f8adb2b7f51d6a7685618a7fb121: We should only
check the version of the binaries in the target installation. The
source installation can of course be of a different version.
Reviewed-by: Daniel Gustafsson <daniel@yesql.se>
Discussion: https://www.postgresql.org/message-id/flat/E1lHNKN-0005IC-V6%40gemulon.postgresql.org
Protocol version 3 was introduced in PostgreSQL 7.4. There shouldn't be
many clients or servers left out there without version 3 support. But as
a courtesy, I kept just enough of the old protocol support that we can
still send the "unsupported protocol version" error in v2 format, so that
old clients can display the message properly. Likewise, libpq still
understands v2 ErrorResponse messages when establishing a connection.
The impetus to do this now is that I'm working on a patch to COPY
FROM, to always prefetch some data. We cannot do that safely with the
old protocol, because it requires parsing the input one byte at a time
to detect the end-of-copy marker.
Reviewed-by: Tom Lane, Alvaro Herrera, John Naylor
Discussion: https://www.postgresql.org/message-id/9ec25819-0a8a-d51a-17dc-4150bb3cca3b%40iki.fi
Feed the desired command to psql via "-c" not stdin, else Perl
may complain that it can't push stdin to an already-failed psql
process, as seen in intermittent buildfarm failures.
Make some minor cosmetic improvements while at it.
Before commit ee28cacf6 we had no tests here that expected failure
to connect, so there seems no need for a back-patch.
Discussion: https://postgr.es/m/CALDaNm2mo8YED=M2ZJKGf1U3F3mw6SaQuLXWCK8rZP6sECYcrA@mail.gmail.com
This expands the binary validation in pg_upgrade with a version
check per binary to ensure that the target cluster installation
only contains binaries from the target version.
In order to reduce duplication, validate_exec is exported from
port.h and the local copy in pg_upgrade is removed.
Author: Daniel Gustafsson <daniel@yesql.se>
Discussion: https://www.postgresql.org/message-id/flat/9328.1552952117@sss.pgh.pa.us
Commit 0aa8a01d04 extends the output plugin API to allow decoding of
prepared xacts and allowed the user to enable/disable the two-phase option
via pg_logical_slot_get_changes(). This can lead to a problem such that
the first time when it gets changes via pg_logical_slot_get_changes()
without two_phase option enabled it will not get the prepared even though
prepare is after consistent snapshot. Now next time during getting changes,
if the two_phase option is enabled it can skip prepare because by that
time start decoding point has been moved. So the user will only get commit
prepared.
Allow to enable/disable this option at the create slot time and default
will be false. It will break the existing slots which is fine in a major
release.
Author: Ajin Cherian
Reviewed-by: Amit Kapila and Vignesh C
Discussion: https://postgr.es/m/d0f60d60-133d-bf8d-bd70-47784d8fabf3@enterprisedb.com
In addition to the existing options of "any" and "read-write", we
now support "read-only", "primary", "standby", and "prefer-standby".
"read-write" retains its previous meaning of "transactions are
read-write by default", and "read-only" inverts that. The other
three modes test specifically for hot-standby status, which is not
quite the same thing. (Setting default_transaction_read_only on
a primary server renders it read-only to this logic, but not a
standby.)
Furthermore, if talking to a v14 or later server, no extra network
round trip is needed to detect the session's status; the GUC_REPORT
variables delivered by the server are enough. When talking to an
older server, a SHOW or SELECT query is issued to detect session
read-only-ness or server hot-standby state, as needed.
Haribabu Kommi, Greg Nancarrow, Vignesh C, Tom Lane; reviewed at
various times by Laurenz Albe, Takayuki Tsunakawa, Peter Smith.
Discussion: https://postgr.es/m/CAF3+xM+8-ztOkaV9gHiJ3wfgENTq97QcjXQt+rbFQ6F7oNzt9A@mail.gmail.com
This option provides REINDEX (TABLESPACE) for reindexdb, applying the
tablespace value given by the caller to all the REINDEX queries
generated.
While on it, this commit adds some tests for REINDEX TABLESPACE, with
and without CONCURRENTLY, when run on toast indexes and tables. Such
operations are not allowed, and toast relation names are not stable
enough to be part of the main regression test suite (even if using a PL
function with a TRY/CATCH logic, as CONCURRENTLY could not be tested).
Author: Michael Paquier
Reviewed-by: Mark Dilger, Daniel Gustafsson
Discussion: https://postgr.es/m/YDiaDMnzLICqeukl@paquier.xyz
Adjust some "can't happen" error messages that assumed that the page
deletion target page must be a half-dead page. This assumption was
wrong in the case of an internal target page. Simply refer to these
pages as the target page instead.
Internal pages are never marked half-dead. There is exactly one
half-dead page for each subtree undergoing deletion. The half-dead page
is also the target subtree's leaf-level page. This has been the case
since commit efada2b8, which totally overhauled nbtree page deletion.
This allows clients to find out the setting at connection time without
having to expend a query round trip to do so; which is helpful when
trying to identify read/write servers. (One must also look at
in_hot_standby, but that's already GUC_REPORT, cf bf8a662c9.)
Modifying libpq to make use of this will come soon, but I felt it
cleaner to push the server change separately.
Haribabu Kommi, Greg Nancarrow, Vignesh C; reviewed at various times
by Laurenz Albe, Takayuki Tsunakawa, Peter Smith.
Discussion: https://postgr.es/m/CAF3+xM+8-ztOkaV9gHiJ3wfgENTq97QcjXQt+rbFQ6F7oNzt9A@mail.gmail.com
On Windows, CMD.EXE allegedly does not run a command that uses forward slashes,
so let's convert the path to use backslashes instead.
Backpatch to 10.
Author: Nitin Jadhav <nitinjadhavpostgres@gmail.com>
Reviewed-by: Juan José Santamaría Flecha <juanjo.santamaria@gmail.com>
Discussion: https://postgr.es/m/CAMm1aWaNDuaPYFYMAqDeJrZmPtNvLcJRS++CcZWY8LT6KcoBZw@mail.gmail.com
This extends the changes made in commit cebc1d34e, teaching
parseqatom() to generate fewer or cheaper subre nodes in some edge
cases. The case of interest here is a quantified atom that is "messy"
only because it has greediness opposite to what preceded it (whereas
captures and backrefs are intrinsically messy). In this case we don't
need an iteration node, since we don't care where the sub-matches of
the quantifier are; and we might also not need a second concatenation
node. This seems of only marginal real-world use according to my
testing, but I wanted to get it in before wrapping up this series of
regex performance fixes.
Discussion: https://postgr.es/m/1340281.1613018383@sss.pgh.pa.us
In some cases, at the time that we're doing an NFA-based precheck
of whether a backref subexpression can match at a particular place
in the string, we already know which substring the referenced
subexpression matched. If so, we might as well forget about the NFA
and just compare the substring; this is faster and it gives an exact
rather than approximate answer.
In general, this optimization can help while we are prechecking within
the second child expression of a concat node, while the capture was
within the first child expression; then the substring was saved during
cdissect() of the first child and will be available to NFA checks done
while cdissect() recurses into the second child. It can help quite a
lot if the tree looks like
concat
/ \
capture concat
/ \
expensive stuff backref
as we will be able to avoid recursively dissecting the "expensive
stuff" before discovering that the backref isn't satisfied with a
particular midpoint that the lower concat node is testing. This
doesn't help if the concat tree is left-deep, as the capture node
won't get set soon enough (and it's hard to fix that without changing
the engine's match behavior). Fortunately, right-deep concat trees
are the common case.
Patch by me, reviewed by Joel Jacobson
Discussion: https://postgr.es/m/661609.1614560029@sss.pgh.pa.us
Bruce Momjian
Thomas Munro
Tom Lane
Peter Eisentraut
Michael Paquier
Amit Kapila
Fujii Masao
Andrew Dunstan
Heikki Linnakangas
Peter Geoghegan
Alvaro Herrera