[cherry release/3.4.x] cherry-picking important fixes to 34x #12403

- fix: lazily initialize structures to avoid c-boundary errors on require [87](Kong/lua-resty-aws#87) - fix: remove more module-level uses of config.global [83](Kong/lua-resty-aws#83) - fix: don't invoke region detection code on the module toplevel and advise against trying to. [81](Kong/lua-resty-aws#81) Signed-off-by: Aapo Talvensaari <aapo.talvensaari@gmail.com>

…n log serialize (#11566) In the function of `kong.log.serialize`, the lifecycle of three table `ctx.route`, `ctx.service` and `ctx.authenticated_consumer` is across request. Modifying the sub-items in the current request of the table will affect the next request, resulting in unexpected behavior in Kong. Fix FTI-5357 --------- Signed-off-by: sabertobihwy <sabertobihwy@gmail.com>

…11859) * fix(rate-limiting): redis async updates When the periodic sync to redis feature is turned on, using the `sync_rate` configuration option, keys are incremented by steps of 2 instead of 1 for requests that arrive after the `sync_rate` interval has expired. This happens because after each sync, the key is loaded again from redis and also incremented atomically (see: #10559) however the next call to `increment` also adds 1 to its value, so the key is incremented by 2 every time it's loaded from redis. This fix sets a negative delta for the key when `conf.sync_rate ~= SYNC_RATE_REALTIME` and the key was loaded from redis in order to invalidate the next call to `increment`. Includes a small code refactor

All rate-limiting plugin instance syncs with the same plugin config, that is the very first config got hit by a request, and they all sync with the same rate. Even a config update won't change the DB to be synced. The timer will sync not just the same instance's counters but all counters in the same DB. This is a compromise given the emergency and we prefer simplicity over correctness for this behavior. Full changelog - The counter table is split with DB; - Timers are created when a request hits; - The sync_rate is guaranteed with limited running timers and timer delay - Cover the case in the integration test by "with_sync_rate" Fix KAG-2904 Co-authored-by: samugi <samuele@konghq.com>

### Summary The PR #11480 introduced a bug that calls `store_connection` without passing `self`. This fixes that. Signed-off-by: Aapo Talvensaari <aapo.talvensaari@gmail.com>

…11464) * fix(declarative): fix TTL not working in DB-less and Hybrid mode * Previously, in DB-less and Hybrid mode, the ttl/updated_at fields were not copied from the original entities to the flattened entities. As a result, the entities were loaded without the TTL field. * Additionally, for loading the TTL field, the "off" DB strategy (lmdb) did not properly filter expired items, nor returned right TTL value for DAO. FTI-4512 * fix coding style * fix coding style: improved function name * added test case: hybrid mode for key-auth * fix test case warnings * fixed test case consumer domain * export ttl as absolute value * delete unused defination * move ttl-fixing logic into row_to_entity() * still use pg to caculate relative value * clean code * add changelog entry * fixed test cases * fixed test cases warning * fixed test failure * fix test case issue: ttl expiration * fix test case: unsed local variable * add an entry in CHANGELOG.md * fix changelog scope * remove release-related information in CHANGELOG.md * fix test case: sleep before attempting unnecessary requests * sleep before attempting unnecessary requests * decrease the ttl to expedite the case's execution * fix CHANGELOG typo * fix the tense problem of changelog entry * add export options for "page_*_for_export" sql statement * fix warning: setting non-standard global variable * fix error reporting: options is nil * fix an issue where the off strategy returned the expired entity * run ttl processing before schema:process_auto_fields()

kong acl plugin will need to get all consumer groups by customer id (because consumer group and acl plugin config some complex connection), these entities can not be warmed up by Kong, The "warm up" mechanism mainly ensures that all entities are loaded into the cache and enables more precise SQL queries (based on cache key definitions) https://github.com/Kong/kong/blob/master/kong/plugins/acl/daos.lua#L8. However, for scenarios where it is necessary to iterate through all entities under a consumer ID, caching is not possible. https://github.com/Kong/kong/blob/master/kong/plugins/acl/groups.lua#L45 So this PR tries a patch warmup mechanism to support acls entities config. FTI-5275

…using kong.response function (#11437) When kong uses kong_buffered_http and upstream timeout, the return status code needs to return 504 (which return by ngx.location.capture), not 502. FTI-5320

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[cherry release/3.4.x] cherry-picking important fixes to 34x #12403

[cherry release/3.4.x] cherry-picking important fixes to 34x #12403

Commits on Mar 7, 2024