underlying catalog not only the index itself. Otherwise, if the cache
load process touches the catalog (which will happen for many though not
all of these indexes), we are locking index before parent table, which can
result in a deadlock against processes that are trying to lock them in the
normal order. Per today's failure on buildfarm member gothic_moth; it's
surprising the problem hadn't been identified before.
Back-patch to 8.2. Earlier releases didn't have the issue because they
didn't try to lock these indexes during load (instead assuming that they
couldn't change schema at all during multiuser operation).
occurring during a reload, such as query-cancel. Instead of zeroing out
an existing relcache entry and rebuilding it in place, build a new relcache
entry, then swap its contents with the old one, then free the new entry.
This avoids problems with code believing that a previously obtained pointer
to a cache entry must still reference a valid entry, as seen in recent
failures on buildfarm member jaguar. (jaguar is using CLOBBER_CACHE_ALWAYS
which raises the probability of failure substantially, but the problem
could occur in the field without that.) The previous design was okay
when it was made, but subtransactions and the ResourceOwner mechanism
make it unsafe now.
Also, make more use of the already existing rd_isvalid flag, so that we
remember that the entry requires rebuilding even if the first attempt fails.
Back-patch as far as 8.2. Prior versions have enough issues around relcache
reload anyway (due to inadequate locking) that fixing this one doesn't seem
worthwhile.
possibility of shared-inval messages causing a relcache flush while it tries
to fill in missing data in preloaded relcache entries. There are actually
two distinct failure modes here:
1. The flush could delete the next-to-be-processed cache entry, causing
the subsequent hash_seq_search calls to go off into the weeds. This is
the problem reported by Michael Brown, and I believe it also accounts
for bug #5074. The simplest fix is to restart the hashtable scan after
we've read any new data from the catalogs. It appears that pre-8.4
branches have not suffered from this failure, because by chance there were
no other catalogs sharing the same hash chains with the catalogs that
RelationCacheInitializePhase2 had work to do for. However that's obviously
pretty fragile, and it seems possible that derivative versions with
additional system catalogs might be vulnerable, so I'm back-patching this
part of the fix anyway.
2. The flush could delete the *current* cache entry, in which case the
pointer to the newly-loaded data would end up being stored into an
already-deleted Relation struct. As long as it was still deleted, the only
consequence would be some leaked space in CacheMemoryContext. But it seems
possible that the Relation struct could already have been recycled, in
which case this represents a hard-to-reproduce clobber of cached data
structures, with unforeseeable consequences. The fix here is to pin the
entry while we work on it.
In passing, also change RelationCacheInitializePhase2 to Assert that
formrdesc() set up the relation's cached TupleDesc (rd_att) with the
correct type OID and hasoids values. This is more appropriate than
silently updating the values, because the original tupdesc might already
have been copied into the catcache. However this part of the patch is
not in HEAD because it fails due to some questionable recent changes in
formrdesc :-(. That will be cleaned up in a subsequent patch.
it is trying to build a relcache entry for. This is an oversight in my 8.2
patch that tried to ensure we always took a lock on a relation before trying
to build its relcache entry. The implication is that if someone committed a
reindex of a critical system index at about the same time that some other
backend were starting up without a valid pg_internal.init file, the second one
might PANIC due to not seeing any valid version of the index's pg_class row.
Improbable case, but definitely not impossible.
caches that we don't actually need to touch. This saves some trivial
number of cycles and avoids certain cases of deadlock when doing concurrent
VACUUM FULL on system catalogs. Per report from Gavin Roy.
Backpatch to 8.2. In earlier versions, CatalogCacheInitializeCache didn't
lock the relation so there's no deadlock risk (though that certainly had
plenty of risks of its own).
it affects. The original coding neglected tablespace entirely (causing
the indexes to move to the database's default tablespace) and for an index
belonging to a UNIQUE or PRIMARY KEY constraint, it would actually try to
assign the parent table's reloptions to the index :-(. Per bug #3672 and
subsequent investigation.
8.0 and 8.1 did not have reloptions, but the tablespace bug is present.
stale relcache init files (pg_internal.init), and there is no mechanism for
updating them during WAL replay. Easiest solution is just to delete the init
files at conclusion of startup, and let the first backend started in each
database take care of rebuilding the init file. Simon Riggs and Tom Lane.
Back-patched to 8.1. Arguably this should be fixed in 8.0 too, but it would
require significantly more code since 8.0 has no handy startup-time scan of
pg_database to piggyback on. Manual solution of the problem is possible
in 8.0 (just delete the pg_internal.init files before starting WAL replay),
so that may be a sufficient answer.
been initialized yet. This can happen because there are code paths that call
SysCacheGetAttr() on a tuple originally fetched from a different syscache
(hopefully on the same catalog) than the one specified in the call. It
doesn't seem useful or robust to try to prevent that from happening, so just
improve the function to cope instead. Per bug#2678 from Jeff Trout. The
specific example shown by Jeff is new in 8.1, but to be on the safe side
I'm backpatching 8.0 as well. We could patch 7.x similarly but I think
that's probably overkill, given the lack of evidence of old bugs of this ilk.
the SQL spec, viz IS NULL is true if all the row's fields are null, IS NOT
NULL is true if all the row's fields are not null. The former coding got
this right for a limited number of cases with IS NULL (ie, those where it
could disassemble a ROW constructor at parse time), but was entirely wrong
for IS NOT NULL. Per report from Teodor.
I desisted from changing the behavior for arrays, since on closer inspection
it's not clear that there's any support for that in the SQL spec. This
probably needs more consideration.
can create or modify rules for the table. Do setRuleCheckAsUser() while
loading rules into the relcache, rather than when defining a rule. This
ensures that permission checks for tables referenced in a rule are done with
respect to the current owner of the rule's table, whereas formerly ALTER TABLE
OWNER would fail to update the permission checking for associated rules.
Removal of separate RULE privilege is needed to prevent various scenarios
in which a grantee of RULE privilege could effectively have any privilege
of the table owner. For backwards compatibility, GRANT/REVOKE RULE is still
accepted, but it doesn't do anything. Per discussion here:
http://archives.postgresql.org/pgsql-hackers/2006-04/msg01138.php
internal TypInfo table in bootstrap mode. This allows array_in and
array_out to be used during early bootstrap, which eliminates the
former obstacle to giving OUT parameters to built-in functions.
(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.
discussion (including making def_arg allow reserved words), add missed
opt_definition for UNIQUE case. Put the reloptions support code in a less
random place (I chose to make a new file access/common/reloptions.c).
Eliminate header inclusion creep. Make the index options functions safely
user-callable (seems like client apps might like to be able to test validity
of options before trying to make an index). Reduce overhead for normal case
with no options by allowing rd_options to be NULL. Fix some unmaintainably
klugy code, including getting rid of Natts_pg_class_fixed at long last.
Some stylistic cleanup too, and pay attention to keeping comments in sync
with code.
Documentation still needs work, though I did fix the omissions in
catalogs.sgml and indexam.sgml.
by creating a reference-count mechanism, similar to what we did a long time
ago for catcache entries. The back branches have an ugly solution involving
lots of extra copies, but this way is more efficient. Reference counting is
only applied to tupdescs that are actually in caches --- there seems no need
to use it for tupdescs that are generated in the executor, since they'll go
away during plan shutdown by virtue of being in the per-query memory context.
Neil Conway and Tom Lane
remove the infrastructure needed to enforce the limit, ie, the global
LRU list of cache entries. On small-to-middling databases this wins
because maintaining the LRU list is a waste of time. On large databases
this wins because it's better to keep more cache entries (we assume
such users can afford to use some more per-backend memory than was
contemplated in the Berkeley-era catcache design). This provides a
noticeable improvement in the speed of psql \d on a 10000-table
database, though it doesn't make it instantaneous.
While at it, use per-catcache settings for the number of hash buckets
per catcache, rather than the former one-size-fits-all value. It's a
bit silly to be using the same number of hash buckets for, eg, pg_am
and pg_attribute. The specific values I used might need some tuning,
but they seem to be in the right ballpark based on CATCACHE_STATS
results from the standard regression tests.
it's not necessary to have three separate calls anymore. This patch also
fixes things so we don't try to read pg_internal.init until after we've
obtained lock on the target database; which was fairly harmless, but it's
certainly cleaner this way.
in various places that were previously doing ad hoc pg_database searches.
This may speed up database-related privilege checks a little bit, but
the main motivation is to eliminate the performance reason for having
ReverifyMyDatabase do such a lot of stuff (viz, avoiding repeat scans
of pg_database during backend startup). The locking reason for having
that routine is about to go away, and it'd be good to have the option
to break it up.
thereby saving a visit to the metapage in most index searches/updates.
This wouldn't actually save any I/O (since in the old regime the metapage
generally stayed in cache anyway), but it does provide a useful decrease
in bufmgr traffic in high-contention scenarios. Per my recent proposal.
that apply the necessary domain constraint checks immediately. This fixes
cases where domain constraints went unchecked for statement parameters,
PL function local variables and results, etc. We can also eliminate existing
special cases for domains in places that had gotten it right, eg COPY.
Also, allow domains over domains (base of a domain is another domain type).
This almost worked before, but was disallowed because the original patch
hadn't gotten it quite right.
functions are not strict, they will be called (passing a NULL first parameter)
during any attempt to input a NULL value of their datatype. Currently, all
our input functions are strict and so this commit does not change any
behavior. However, this will make it possible to build domain input functions
that centralize checking of domain constraints, thereby closing numerous holes
in our domain support, as per previous discussion.
While at it, I took the opportunity to introduce convenience functions
InputFunctionCall, OutputFunctionCall, etc to use in code that calls I/O
functions. This eliminates a lot of grotty-looking casts, but the main
motivation is to make it easier to grep for these places if we ever need
to touch them again.
regardless of the current schema search path. Since CREATE OPERATOR CLASS
only allows one default opclass per datatype regardless of schemas, this
should have minimal impact, and it fixes problems with failure to find a
desired opclass while restoring dump files. Per discussion at
http://archives.postgresql.org/pgsql-hackers/2006-02/msg00284.php.
Remove now-redundant-or-unused code in typcache.c and namespace.c,
and backpatch as far as 8.0.
index's support-function cache (in index_getprocinfo). Since none of that
data can change for an index that's in active use, it seems sufficient to
treat all open indexes the same way we were treating "nailed" system indexes
--- that is, just re-read the pg_class row and leave the rest of the relcache
entry strictly alone. The pg_class re-read might not be strictly necessary
either, but since the reltablespace and relfilenode can change in normal
operation it seems safest to do it. (We don't support changing any of the
other info about an index at all, at the moment.)
Back-patch as far as 8.0. It might be possible to adapt the patch to 7.4,
but it would take more work than I care to expend for such a low-probability
problem. 7.3 is out of luck for sure.
This is utterly insignificant in normal operation, but it becomes a
problem during cache inval stress testing. The original coding in fact
had no leak --- the 8.0 List rewrite created the issue. I wonder whether
list_concat should pfree the discarded header?
and nail a couple more system indexes into cache. This doesn't make
any difference in normal system operation, but when forcing constant
cache resets it's difficult to get through the rules regression test
without these changes.
dead and have become unreferenced. Before 8.1, such members were left
for AtEOXact_CatCache() to clean up, but now AtEOXact_CatCache isn't
supposed to have anything to do. In an assert-enabled build this bug
leads to an assertion failure at transaction end, but in a non-assert
build the dead member is effectively just a small memory leak.
Per report from Jeremy Drake.
(previously we only did = and <> correctly). Also, allow row comparisons
with any operators that are in btree opclasses, not only those with these
specific names. This gets rid of a whole lot of indefensible assumptions
about the behavior of particular operators based on their names ... though
it's still true that IN and NOT IN expand to "= ANY". The patch adds a
RowCompareExpr expression node type, and makes some changes in the
representation of ANY/ALL/ROWCOMPARE SubLinks so that they can share code
with RowCompareExpr.
I have not yet done anything about making RowCompareExpr an indexable
operator, but will look at that soon.
initdb forced due to changes in stored rules.
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.
functionality, but I still need to make another pass looking at places
that incidentally use arrays (such as ACL manipulation) to make sure they
are null-safe. Contrib needs work too.
I have not changed the behaviors that are still under discussion about
array comparison and what to do with lower bounds.
valid type information if they are asked to fetch the values part of a
pg_statistic slot; these arguments are unneeded if fetching only the
numbers part. Use this to save a catcache lookup in btcostestimate,
which is looking like a bit of a hotspot in recent profiling. Not a
big savings, but since it's essentially free, might as well do it.
the parent table, even if the command that creates them is executed by
someone else (such as a superuser or a member of the owning role).
Per gripe from Michael Fuhr.
SearchCatCacheList and ReleaseCatCacheList. Previously, we incremented
and decremented the refcounts of list member tuples along with the list
itself, but that's unnecessary, and very expensive when the list is big.
It's cheaper to change only the list refcount. When we are considering
deleting a cache entry, we have to check not only its own refcount but
its parent list's ... but it's easy to arrange the code so that this
check is not made in any commonly-used paths, so the cost is really nil.
The bigger gain though is to refrain from DLMoveToFront'ing each individual
member tuple each time the list is referenced. To keep some semblance
of fair space management, lists are just marked as used or not since the
last cache cleanout search, and we do a MoveToFront pass only when about
to run a cleanout. In combination, these changes reduce the costs of
SearchCatCacheList and ReleaseCatCacheList from about 4.5% of pgbench
runtime to under 1%, according to my gprof results.
whenever we generate a new OID. This prevents occasional duplicate-OID
errors that can otherwise occur once the OID counter has wrapped around.
Duplicate relfilenode values are also checked for when creating new
physical files. Per my recent proposal.
