Track the insert ID value in Database, similar to the affected rows.
This makes it possible for subclasses to stash or override the value,
which is useful when emulating a write operation using multiple queries.
This includes the case of internal use of atomic sections, where the
COMMIT/RELEASE can reset the last_insert_id tracked in the PECL driver
itself.
Use separate methods and fields for "last query statement" information
and "last query method" information.
Make insertId() for SQLite and Postgres better match MySQL:
* Return 0 if the last query statement did not change any rows.
This helps protect against callers that fail to check affectedRows().
* Make it return the existing ROWID/SERIAL column when upsert() updates
an existing row. This adds a new getInsertIdColumnForUpsert() helper
function.
Directly use query() in doReplace() and doInsertSelectGeneric() to make
the affected row/ID logic easier to follow.
Improve insertId() and affectedRows() documentation.
Add more integration tests of row insertion methods.
Bug: T314100
Change-Id: I7d43a2e52260e66acb713554bb883f5f4a14d010
Subclasses no longer have to implement fetchAffectedRowCount()
and affectedRows() no longer depends on the driver connection
handle.
Set "port" to $wgDBport in LBFactoryTest/LoadBalancerTest to
avoid postgres failures.
Bug: T314100
Change-Id: Ib31a9d2db18d7ba7dcf61fb110d0fef53f455464
This is part of LB doing wearing way too many hats. LB's job is to take
an index out of a hat, not to init the db object. We have a factory for
that already.
Also completely get rid of injecting DBFactory as a parameter in config
of LB and LBF, It doesn't make any sense to pass this around as a
configuration option, it's a php class, so can't even be properly set in
many systems (code should not be a configuration). On top of that it's
making multiple ways to override configurations with non-obvious
priority that can easily lead to outages.
Bug: T326274
Change-Id: I1e0c38cd3b378669d0940b9f243b61cb64c193b7
Follows-up I4c7d826c7ec654b, I1287f3979aba1bf1.
We lose useful coverage and spend valuable time keeping these accurate
through refactors (or worse, forget to do so). The theoretically "bad"
accidental coverage is almost never actually bad.
Having said that, I'm not removing them wholesale (yet). I've audited
each of these specific files to confirm it is a general test of the
specified subject class, and also kept it limited to those specified
classes. That's imho more than 100% of the benefit for less than 1%
of the cost (more because `@covers` is more valuable than the fragile
and corrosive individual private method tracking in tests that
inevitably get out of date with no local incentive to keep them up to
date).
Cases like structure tests keep `@coversNothing` etc and we still don't
count coverage of other classes. There may be a handful of large
legacy classes where some methods are effectively class-like in
complexity and that's why it's good for PHPUnit to offer the precision
instrument but that doesn't meant we have to use that by-default for
everything.
I think best practice is to write good narrow unit tests, that reflect
how the code should be used in practice. Not to write bad tests and
hide part of its coverage within the same class or even namespace.
Fortunately, that's generally what we do already it's just that we
also kept these annotations still in many cases.
This wastes time to keep methods in sync, time to realize (and fix)
when other people inevitably didn't keep them in sync, time to find
uncovered code only to realize it is already covered, time for a less
experienced engineer to feel obligate to and do write a low quality
test to cover the "missing" branch in an unrealistic way, time wasted
in on-boarding by using such "bad" tests as example for how to use
the code and then having to unlearn it months/years later, loss of
telemetry in knowing what code actually isn't propertly tested due to
being masked by a bad test, and lost oppertunities to find actually
ununused/unreachable code and to think about how to instead structure
the code such that maybe that code can be removed.
------
Especially cases like LBFactoryTest.php were getting out of hand,
and in GlobalIdGeneratorTest.php we even resorted to reminding people
with inline comments to keep tags in sync.
Change-Id: I69b5385868cc6b451e5f2ebec9539694968bf58c
Use server names to handle the case where server indexes get
shifted around due to the depooling or provisioning of a server.
Previously, loads could be assigned to a wrong server, assigned
to a server that was depooled, or assigned to a server that is
to new to appear in the current "servers" array. This would mean
that getReaderIndex() or getLagTimes() could trigger exceptions.
Only change loads for depooled servers. Update the loads of other
servers to the current loads only makes sense in the context of
using all the current servers. This does not apply to reconfigure()
since it does not see newly pooled servers (for simplicity).
Handle servers depooled only from custom query groups.
Bug: T322156
Change-Id: I9f710aa32f5d5b74796bb80a8426a5f653b8e4d3
These options only support one value in the array: maxWriteDuration
Instead of turning this into a full array and pass it around, just pass
the integer to simplify the logic, avoid mistakes by typos and more.
Bug: T326274
Depends-On: Ib5c76346d0a61c3a6906365b3ced9fca2d43e4d2
Change-Id: Ib60f25ba4a7ca1d14d062d9121fe34e94ccc3b70
Factor out normalizeServerMaps() method to simplify getServerName()
by pre-setting "serverName". Add uniqueness safety check, since we
rely on this property in cache keys and in reconfigure().
Change-Id: I06672885c79611e1257adb5f3dec88194e71b705
Notable changes:
* In SqlBagOStuff::getConnectionFromServerInfo, only two loggers were
injected. The rest implicitly got a NullLogger due to being absent.
These are now effectively unsilenced.
* Database::__construct() required almost all parameters, even the
loggers. I've wanted to move some of DatabaseFactory into the ctor
here for a while. In order to make this change not a breaking
change, the new 'logger' parameter is optional with NullLogger as
default. This allowed some of the test cases, which were simply
passing NullLogger, to be fixed by passing nothing instead of
passing the new option name.
The Database class is behind a dozen layers of indirection for
real use, so this will still be injected just fine (DBF, LB, LBF,
MWLBF, etc.).
* In LegacyLogger, the handling for $wgDBerrorLog was previously
limited to DBConnection and DBQuery. This now includes errors
from other (generally, newer) parts of Rdbms as well, which were
previously missing.
This only affects sites (typically CI and dev setup) where
$wgDBerrorLog is used, as opposed to the more common
$wgDebugLogGroups by-channel configuration.
* TransactionProfiler gets its logger injected in a rather odd way,
via entrypoints (MediaWiki.php, ApiMain.php, and MaintenanceRunner)
as opposed to service wiring. This is kept as-is for now.
* In LBFactoryTest, in particular testInvalidSelectDBIndependent2,
there are cases that intentionally produce failures of which
the result is then observed. In CI we assert that dberror.log is
empty so instead of adding the missing logger fields to that
LBFactory instance, the only one set (replLogger) is removed.
The alternative is to set 'logger' now, which would naturally
cause CI failures due to unexpected entries coming through to
non-mocked error log.
Bug: T320873
Change-Id: I7ca996618e41b93f488cb5c4de82000bb36e0dd3
Rename lazyLoadReplicationPositions() and doWait(). The later now
handles the logic of checking if a position is set or if the reader
index is that of the primary DB.
Narrow the job of reallyOpenConnection() by removing "waitForPos"
loading logic. It now only deals with initializing a connection and
not the logic that goes along with tracking/pooled connection handles.
Note that getReaderIndex() still triggers primary position loading and
was the only method that automatically triggered logic to actually wait
on the primary position.
Add more comments to getReaderIndex() and optimize call order slightly.
Minor cleanups to related code comments.
Change-Id: I685b1f7946aadb5463e8870edda4340be3fc4ae2
Remove 'insertSelectIsSafe' option, unused.
Remove 'topologicalPrimaryConnRef' option, no longer used as of two
months ago with I41a57247503 (8c9398f7f9).
Remove unneeded DatabaseSqlite::getTopologyBasedServerId
implementation which can inherit null instead of overriding with string
of "0". Only caller is SqlBagOStuff::makeTimestampedModificationToken
which can be used as MainStash DB, where its important that a given
server always has the same unique name within a set of db hosts that
may replicate to each other. By inheriting null as topology server ID,
it SqlBagOStuff will use IDatabase::getServerName instead. That in turn
uses the 'host' connection parameter, which defaults to null in
DatabaseFactory, and then falls back to the string "unknown" which is
as good as "0" for this purpose.
Bug: T299691
Change-Id: Iceb65c28cdd3c4a89b3c8b34c3f95d3285718ec0
Make DBConnRef enforce the DB domain selected during its lifetime
and allow more nested and successive use of the same connection handle
via DBConnRef. This can avoid extra connections in some cases where
getConnection()/getConnectionRef() is used.
Also:
* Reduce the number of connection pools arrays from six to two
* Merge getLocalConnection()/getForeignConnection() into one method
* Expand various related code comments
Since LoadBalancer::getReadOnlyReason() no longer user the local domain
but rather DOMAIN_ANY, it should not result in "USE" errors if the local
domain does not have a database on the server.
This version of the patch removes the unused reuseConnectionInternal()
method (the method was previously added back to the patch by mistake).
Bug: T226595
Change-Id: I62502f4de4f86a54f25be1699c4d1a1c1baee60b
Make DBConnRef enforce the DB domain selected during its lifetime
and allow more nested and successive use of the same connection handle
via DBConnRef. This can avoid extra connections in some cases where
getConnection()/getConnectionRef() is used.
Also:
* Reduce the number of connection pools arrays from six to two
* Merge getLocalConnection()/getForeignConnection() into one method
* Expand various related code comments
Bug: T226595
Depends-On: If808cbab429d41e1f2289683533e4a781a4bdf5e
Change-Id: I540b08920997c57cad6445ddb09d8e663eaf4714
This method is unused and the naming is not very clear in terms of what
it returns (server name vs config map). Removing it reduces externally
exposed complexity.
Optimize DBConnRef::getServerName() when the server index is known.
Use this to eliminate the "topologicalMaster" parameter from Database.
Rename internal fields and paramters in Database to use "primary"
instead of "master", for consistency.
Add some additional clarifying comments.
Bug: T299691
Change-Id: I98515fa02a58a4c72a06f1ff283b249b1617c886
The 'watchlist' query group no longer exists in WMF production since
T263127, and it's not included in the list of supported query groups
in docs/database.md.
Change-Id: I6ea5c65921891ac6a705a6ff7e79b08fa5a9bf42
This introduces $wgLBFactoryConf['configCallback'] which can be set to a
function that returns updates to be applied to $wgLBFactoryConf. The new
method LBFactory::autoreConfigure() can be called to check the callabck
and, if the config changed, reconfigure all existing LoadBalancers.
Reconfiguring the LoadBalancers causes all open connections to be
invalidated; however, any DBConnRef instances will remain valid and will
acquire a fresh connection from the LoadBalancer automatically when
appropriate.
As a proof of concept, this patch adds support for config reloding
into WikiExporter.
Bug: T298485
Change-Id: I6c3ffde62f6e038730736abe980befd90ec43e1a
This would simplify the code for its users a lot.
Bug: T255493
Depends-On: I6ab4440a13f4682dd3a6e9cfc6c629222fbeab8a
Change-Id: I6e7544763bde56fc1e19de0358b71ba984a978ca
Callback style iteration made sense before generators existed, but
generators make for simpler code. The "call method" variants made
sense before closures existed but defeat static analysis.
So, in LBFactory:
* Add ILBFactory::getAllLBs()
* Deprecate ILBFactory::forEachLB()
* Remove LBFactory::forEachLBCallMethod(), was protected.
* Add LBFactory::getLBsForOwner(), which is protected and has the
internal interface in @return. Adding a new abstract method breaks
Wikibase tests despite LBFactory not being stable to extend.
* Migrate callers. Generators allow you to return/break from the middle
of the loop, which implies a little rearrangement for some callers.
In LoadBalancer, connections supposedly of type IDatabase were
routinely type-hinted as Database in closure parameters so that methods
could be called that were not in the interface. So it's convenient to
get rid of public iteration methods entirely in favour of private
methods returning Database[].
* Hard-deprecate ILoadBalancerForOwner::forEachOpenConnection()
and replace it with a private generator method since nothing called
it externally except for core tests.
* Hard-deprecate ILoadBalancerForOwner::forEachOpenPrimaryConnection()
and replace it with a private generator. DeferredUpdates needed it for
iterating over IDatabase::explicitTrxActive(), so add
ILoadBalancer::explicitTrxActive() as a replacement.
* Replace private method LoadBalancer::forEachOpenReplicaConnection()
with a generator.
Depends-On: If0b382231e27d6d1197fb7b6aef6ab50335df4e5
Change-Id: I64514e77b9bfe737be5b12e1d3c9c49976bb522f
Rename canRecoverFromDisconnect() in order to better describe
its function. Make it use the transaction ID and query walltime
as arguments and return an ERR_* class constant instead of a bool.
Avoid retries of slow queries that yield lost connection errors.
Track session state errors caused by the loss of named locks or
temp tables (e.g. during connection loss). Such errors will prevent
further queries except for rollback() and flushSession(), which must
be issued to resolve the error.
Add flushPrimarySessions() methods to LBFactory/LoadBalancer
and use it in places where session state loss is meant to be
safely aknowledged.
Change-Id: I60532f86e629c83b357d4832d1963eca17752944
This reverts commit eed58f2f61.
Reason for revert: Blocking the train, going to revert the whole chain
Change-Id: I64f9e5a9dde106671783f958a686ca697182077b
Use the LoadBalancer id in flushPrimarySessions(), not the LBFactory one,
and use assertOwnership() to check $owner, similar to other methods.
In DatabaseMysqlBase::doFlushSession(), change RELEASE_ALL_LOCKS() query
to use RELEASE_LOCK(), since only newer MariaDB versions (>=10.5.2) support
it. No errors were thrown in the method since they are suppressed, but the
syntax error would cause the transaction to be placed in an error state.
Add assertion to testTransactionCallbackChains() that would otherwise fail.
Randomize lock names in lock() tests to avoid contention.
Bug: T292239
Bug: T303887
Follow-Up: ee3c65d541
Follow-Up: 4cac31de4e
Change-Id: I414d737028338cfd5369eee24576df4aa26a2f6f
Make getServer() always return a string, as documented, even with new
Database::NEW_UNCONNECTED handles that have yet to call open(). If the
'host' parameter to __construct() is ''/null, getServer() now returns
'localhost' instead of null. This avoids problems like fatal errors in
calls to TransactionProfiler::recordConnection().
Use Database constants for "connectionParams" field keys for better
static analysis.
Also:
* Add Database::getServerName() method that returns "readable" server
names in the style of LoadBalancer::getServerName(). Note that the
"hostName" field is already passed in from LoadBalancer.
* Migrate most getServer() callers to getServerName() for easier
debugging and more readable logging.
* Also, normalize Database/LoadBalancer SPI logging context to use
"db_server" and reduce logging code duplication in LoadBalancer.
Bug: T277056
Change-Id: I00ed4049ebb45edab1ea07561c47e226a423ea3b
Also:
* Add LBFactoryMulti sanity check to avoid section/cluster collisions.
Since sections are just a type of cluster, using the same name can
cause confusion. WMF config already treats them as overall unique.
* Rename 'externalLoads' field in LoadBalancer for clarity.
Change-Id: I748db691b4918b797b82d75cfd8722453ccd9d7f
