XianReallyHot-ZZH opened a new pull request, #13834: URL: https://github.com/apache/ignite/pull/13834
Thank you for submitting the pull request to the Apache Ignite. In order to streamline the review of the contribution we ask you to ensure the following steps have been taken: ### The Contribution Checklist - [x] There is a single JIRA ticket related to the pull request. - [ ] The web-link to the pull request is attached to the JIRA ticket. - [ ] The JIRA ticket has the _Patch Available_ state. - [x] The pull request body describes changes that have been made. The description explains _WHAT_ and _WHY_ was made instead of _HOW_. - [x] The pull request title is treated as the final commit message. The following pattern must be used: `IGNITE-XXXX Change summary` where `XXXX` - number of JIRA issue. - [ ] A reviewer has been mentioned through the JIRA comments (see [the Maintainers list](https://cwiki.apache.org/confluence/display/IGNITE/How+to+Contribute#HowtoContribute-ReviewProcessandMaintainers)) - [ ] The pull request has been checked by the Teamcity Bot and the `green visa` attached to the JIRA ticket (see tab `PR Check` at [TC.Bot - Instance 1](https://tcbot2.sbt-ignite-dev.ru/prs.html) or [TC.Bot - Instance 2](https://mtcga.gridgain.com/prs.html)) ### Description Ticket: [IGNITE-29297](https://issues.apache.org/jira/browse/IGNITE-29297) The group-level SQL onheap index row cache (`IndexRowCache`) is keyed by a bare data row link, but a link is an address, not an identity: after a bulk free that bypasses per-row row-cache cleanup (a partition destroy on rebalancing) and a subsequent link reuse by a row of another table of the same cache group, a stale cached entry of the old owner was served to index operations of the new owner — `InlineIndexTree#createIndexRow` re-wrapped it with the current index's row handler without any ownership validation. Up to 2.17 this fails with `BinaryObjectException` wrapped into `CorruptedTreeException` (node restarts, false "corrupted index" maintenance rebuilds); since 2.18 (IGNITE-26261) the failure is silent: one table's index is maintained with another table's field values, and range scans return foreign rows (reproduced on master: a scan of one table returned 50 rows of another table out of 51). **What**: every cached row is stamped with the cache ID of the index tree that cached it. On lookup, an entry stamped for another cache of the group is evicted and the row is re-read from the link, so a stale entry can no longer be served across caches — regardless of which bulk-free path left it behind. `IndexRowCache#clearForCache` compares the same stamp, which also makes it effective for groups without page eviction (cache IDs are stored in data pages only when page eviction is enabled, so the previously compared `CacheDataRow#cacheId` was always 0 in the default configuration). **Why this scope**: consumer-side validation makes the row cache correct against any staleness source, so a producer-side sweep of the partition destroy path is a possible follow-up, not a correctness requirement; cache destruction is already covered by the per-row cleaner in `RowStore#removeRow` (see `H2RowCacheSelfTest#checkDestroyCacheWithOtherCacheInGroup`). **Testing**: the new `IndexRowCacheCrossTablePoisonTest` injects the exact post-bulk-free state (entries of cache B under links of cache A, the state the node ends up in after link reuse) and verifies that both index operation paths of cache A — range scan (find) and insert (put) — observe only rows of cache A: before the fix the scan returned 50 rows of cache B, after the fix 0. `H2RowCacheSelfTest` (16 tests) and `H2RowCachePageEvictionTest` (4 tests) pass unchanged. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
