* This fixes keys based on some sort of change log.
Updates are wrapped in a mutex and keep track of the
last known good position.
* Make WANObjectReapUpdate class that cleans up title
related keys using the recentchanges table. This triggers
as a deferred updates on RC view.
Change-Id: I7f14b9ca2533032147e62b1a3cc004a23da86579
Causes HHVM 3.12.7 to segfault and or:
Fatal error: Stack overflow in includes/libs/objectcache/BagOStuff.php on line 754
This reverts commit b47ce21cec.
Bug: T150833
Change-Id: Ia52fdf65f5e386c0eaa1046328680fb5c8a9081a
This works by setting a callback to return the cache set
options. The callback will watch DB reads and create a
merged result from said usage.
This handles callers that are missing getCacheSetOptions().
Change-Id: Ia264f011e45e8cf105480955dad7e2c4c2357b73
Because the process cache can be lagged by virtue of blind TTL,
the HOLDOFF_TTL might not be enough to account for it, so avoid
using it when already inside a callback.
Also split of the tests from the MediaWiki test class, so this
does not require DB access anymore.
Change-Id: I743a1233a5efc7f036fad140a9ff8a30b32f8f27
* Renamed mDoneWrites to be clearer at what type it is.
* Also cleaned up a few callers of this method
Change-Id: I45856b210c289c2e2f193cc4328a208e20b4e0a8
This does what it says on the tin, e.g. a way to fetch
multiple keys at once. Callbacks are still called in a
serial manner when needed.
Change-Id: I8e24a6de7f46499a53ec41636c5a4f106b9b3d09
This lets callers use adaptive TTLs on the near-expiration
preemptive refreshes if the new and current values match,
using the as-of time as $mtime.
Change-Id: Ie541c35f890c9f789d1accf9f2a43506daaf31f0
This lets callers have their own process cache but keep it
managed in a central location, making it easier to reset
when DB transaction snapshot are flushed or when unit tests
want to clear caches between runs.
Use this to replace the Revision text process cache.
Change-Id: Ic61ee9140d4ce9836cc4650adb5bb75a291fea18
Callers using wfTimestamp( TS_UNIX, ... ) where getting $minTTL
due to the output being a string number.
Change-Id: I6b67a941940f40ef9a543f11d0dbccacafaaa53b
These control worthRefreshPopular(), which will try to make sure that
heavily used keys are kept more consistent than others.
Change-Id: I2d351b50462a96d0fa8ca8fa6d28d51d3a5c94ff
This is more consistent with LoadBalancer, modern, and inclusive
of master/master mysql, NDB cluster, and MariaDB galera cluster.
The old constant is an alias now.
Change-Id: I0b37299ecb439cc446ffbe8c341365d1eef45849
* This better handles delayed/lost cache purges by
having lower TTLs for entries that often changes.
* Use this for foreign upload description page caches,
we purges are never received from the source wiki.
* Also use this for User and LocalFile cache TTLs.
* Also move the Database::getCacheSetOptions() call in
User *before* doing the queries, which is preferred.
* Fixed some IDEA errors too, like the undeclared
mApiBase field.
Change-Id: I70f8ebb29ac853c2a530d9eedb9e7facc1b7b710
* Update documentation about relay methods.
* Change interim key set() to add() to avoid broadcasting it.
* Remove the behavior of doing purges synchronously in the
local DC first before relay. In both the event relayer and
mcrouter case, they will be asynchronous. It was hardly
even possible to use such behavior since loads come from
slave DBs, which do not see changes right after COMMIT.
Bug: T97562
Change-Id: I7759c82ae6e1b72fc227882a99c9a712a46374f6
* Add the ability to expose key BagOStuff attributes like whether the
emulation SQL cache is being used. Several callers use hacks to detect
this and can be updated later.
* Fallback to a safe empty WAN cache in the CACHE_DB case. This fixes
a regression from f4bf52e843.
* Also add this protection to the server cache, which could break
similarly before too.
* Also fix CACHE_ANYTHING by avoid the recursion risk from
fc1d4d7960 by just checking the disabled service map.
Bug: T123829
Bug: T141804
Change-Id: I17ee26138f69e01ec1aaddb55ab27caa4d542193
This is useful for avoiding stampedes in the one case that lockTSE
does not alone cover, which is when the key does not exist or is
tombstoned.
Also avoid saving interim values unless the key is tombstoned
since there is no point in doing that otherwise.
Change-Id: I70997e90217a0979e0589afa7a5107b0e623c7cf
* getWithSetCallback() takes a 'version' parameter.
* If the value at a key has a different version, then
getWithSetCallback() will automatically use a separate
key. Which value "wins" the main key does not matter.
* Purges are handled by using the main key as a sort of
check key (with no hold-off). Note that this key is always
purged on delete().
* Changed stash keys to track the same info as other keys
both for consistency and because this change needs the
generation timestamp. Renamed the stash prefix to avoid
corrupt results with Het Deploy.
* This is useful for things like the User class that use
versioning and have cross-wiki key access and purges.
Currently, bumps to version must be deployed to all wikis
at once, which this aims to avoid.
Change-Id: I26ae62f116e32b48bcf06bc13f8b9e79ae976745
This makes the channels more explicit and defined in a less
ad-hoc way. Systems like Kafka would prefer explicit channel
definitions anyway, so the channel prefix just obscures things.
Change-Id: I5631eb1b1382083396a0f08904d9273cc92601e8
LocalRepo::checkRedirect() cache image redirect checks in memory to avoid
unnecessary trips to memcached. Same idea as Idbd11637.
Change-Id: Idaa12376382aaa0bd8cb988240ec1da0b688359d
MessageBlobStore class:
* Make logger aware.
* Log an error if json encoding fails.
* Stop using the DB table. WANObjectCache supports everything we need:
- Batch retrieval.
- Invalidate keys with wildcard selects or cascading check keys.
* Update tests slightly since the actual update now happens on-demand as
part of get() instead of within updateMessage().
ResourceLoader class:
* Remove all interaction with the msg_resource table. Remove db table later.
* Refactor code to use a hash of the blob instead of a timestamp.
Timestamps are unreliable and roll over too frequently for message blob store
because there is no authoritative source. The timestamps were inferred based on
when a change is observed. Message overrides from the local wiki have an
explicit update event when the page is edited. All other messages, such as
from MediaWiki core and extensions using LocalisationCache, have a single
timestamp for all messages which rolls over every time the cache is rebuilt.
A hash is deterministic, and won't cause needless invalidation (T102578).
* Remove redundant pre-fetching in makeModuleResponse().
This is already done by preloadModuleInfo() in respond().
* Don't bother storing and retreiving empty "{}" objects.
Instead, detect whether a module's message list is empty at runtime.
ResourceLoaderModule class:
* Make logger aware.
* Log if a module's message blob was not preloaded.
cleanupRemovedModules:
* Now that blobs have a TTL, there's no need to prune old entries.
Bug: T113092
Bug: T92357
Change-Id: Id8c26f41a82597e34013f95294cdc3971a4f52ae
It was already used everywhere as non-static via $this.
This is needed in order to allow MessageBlobStore unit tests
to disable the holdoff via a mock (mocks can't override static methods).
Change-Id: I3aad5b6e780addf1b6ce9de56c81b91f5ab358b2
