[IGNITE-9562] Destroyed cache that resurrected on an old offline node breaks PME

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.5
• Fix Version/s: 2.7.6
• Component/s: cache
• Labels: 2.7.6-rc1
• Release Note: Fixed an issue where an outdated node with a destroyed cache caused the cluster to hang
• Ignite Flags: Docs Required, Release Notes Required
• Assignee: Eduard Shangareev
• Reporter: Pavel Kovalenko
• Created: 12/Sep/18 15:51
• Updated: 06/Apr/20 13:45
• Resolved: 20/Aug/19 11:24

Description

Given:
2 nodes, persistence enabled.
1) Stop 1 node
2) Destroy cache through client
3) Start stopped node

When the stopped node joins to cluster it starts all caches that it has seen before stopping.
If that cache was cluster-widely destroyed it leads to breaking the crash recovery process or PME.

Root cause - we don't start/collect caches from the stopped node on another part of a cluster.

In case of PARTITIONED cache mode that scenario breaks crash recovery:

java.lang.AssertionError: AffinityTopologyVersion [topVer=-1, minorTopVer=0]

	at org.apache.ignite.internal.processors.affinity.GridAffinityAssignmentCache.cachedAffinity(GridAffinityAssignmentCache.java:696)
	at org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopologyImpl.updateLocal(GridDhtPartitionTopologyImpl.java:2449)
	at org.apache.ignite.internal.processors.cache.distributed.dht.GridDhtPartitionTopologyImpl.afterStateRestored(GridDhtPartitionTopologyImpl.java:679)
	at org.apache.ignite.internal.processors.cache.persistence.GridCacheDatabaseSharedManager.restorePartitionStates(GridCacheDatabaseSharedManager.java:2445)
	at org.apache.ignite.internal.processors.cache.persistence.GridCacheDatabaseSharedManager.applyLastUpdates(GridCacheDatabaseSharedManager.java:2321)
	at org.apache.ignite.internal.processors.cache.persistence.GridCacheDatabaseSharedManager.restoreState(GridCacheDatabaseSharedManager.java:1568)
	at org.apache.ignite.internal.processors.cache.persistence.GridCacheDatabaseSharedManager.beforeExchange(GridCacheDatabaseSharedManager.java:1308)
	at org.apache.ignite.internal.processors.cache.distributed.dht.preloader.GridDhtPartitionsExchangeFuture.distributedExchange(GridDhtPartitionsExchangeFuture.java:1255)
	at org.apache.ignite.internal.processors.cache.distributed.dht.preloader.GridDhtPartitionsExchangeFuture.init(GridDhtPartitionsExchangeFuture.java:766)
	at org.apache.ignite.internal.processors.cache.GridCachePartitionExchangeManager$ExchangeWorker.body0(GridCachePartitionExchangeManager.java:2577)
	at org.apache.ignite.internal.processors.cache.GridCachePartitionExchangeManager$ExchangeWorker.body(GridCachePartitionExchangeManager.java:2457)
	at org.apache.ignite.internal.util.worker.GridWorker.run(GridWorker.java:110)
	at java.lang.Thread.run(Thread.java:748)

In case of REPLICATED cache mode that scenario breaks PME coordinator process:

[2018-09-12 18:50:36,407][ERROR][sys-#148%distributed.CacheStopAndRessurectOnOldNodeTest0%][GridCacheIoManager] Failed to process message [senderId=4b6fd0d4-b756-4a9f-90ca-f0ee25100001, messageType=class o.a.i.i.processors.cache.distributed.dht.preloader.GridDhtPartitionsSingleMessage]
java.lang.AssertionError: 3080586
	at org.apache.ignite.internal.processors.cache.GridCachePartitionExchangeManager.clientTopology(GridCachePartitionExchangeManager.java:815)
	at org.apache.ignite.internal.processors.cache.distributed.dht.preloader.GridDhtPartitionsExchangeFuture.updatePartitionSingleMap(GridDhtPartitionsExchangeFuture.java:3621)
	at org.apache.ignite.internal.processors.cache.distributed.dht.preloader.GridDhtPartitionsExchangeFuture.processSingleMessage(GridDhtPartitionsExchangeFuture.java:2439)
	at org.apache.ignite.internal.processors.cache.distributed.dht.preloader.GridDhtPartitionsExchangeFuture.access$100(GridDhtPartitionsExchangeFuture.java:137)
	at org.apache.ignite.internal.processors.cache.distributed.dht.preloader.GridDhtPartitionsExchangeFuture$2.apply(GridDhtPartitionsExchangeFuture.java:2261)
	at org.apache.ignite.internal.processors.cache.distributed.dht.preloader.GridDhtPartitionsExchangeFuture$2.apply(GridDhtPartitionsExchangeFuture.java:2249)
	at org.apache.ignite.internal.util.future.GridFutureAdapter.notifyListener(GridFutureAdapter.java:383)
	at org.apache.ignite.internal.util.future.GridFutureAdapter.listen(GridFutureAdapter.java:353)
	at org.apache.ignite.internal.processors.cache.distributed.dht.preloader.GridDhtPartitionsExchangeFuture.onReceiveSingleMessage(GridDhtPartitionsExchangeFuture.java:2249)
	at org.apache.ignite.internal.processors.cache.GridCachePartitionExchangeManager.processSinglePartitionUpdate(GridCachePartitionExchangeManager.java:1628)
	at org.apache.ignite.internal.processors.cache.GridCachePartitionExchangeManager.access$1100(GridCachePartitionExchangeManager.java:141)
	at org.apache.ignite.internal.processors.cache.GridCachePartitionExchangeManager$2.onMessage(GridCachePartitionExchangeManager.java:368)
	at org.apache.ignite.internal.processors.cache.GridCachePartitionExchangeManager$2.onMessage(GridCachePartitionExchangeManager.java:332)
	at org.apache.ignite.internal.processors.cache.GridCachePartitionExchangeManager$MessageHandler.apply(GridCachePartitionExchangeManager.java:2999)
	at org.apache.ignite.internal.processors.cache.GridCachePartitionExchangeManager$MessageHandler.apply(GridCachePartitionExchangeManager.java:2978)
	at org.apache.ignite.internal.processors.cache.GridCacheIoManager.processMessage(GridCacheIoManager.java:1056)
	at org.apache.ignite.internal.processors.cache.GridCacheIoManager.onMessage0(GridCacheIoManager.java:581)
	at org.apache.ignite.internal.processors.cache.GridCacheIoManager.handleMessage(GridCacheIoManager.java:380)
	at org.apache.ignite.internal.processors.cache.GridCacheIoManager.handleMessage(GridCacheIoManager.java:306)
	at org.apache.ignite.internal.processors.cache.GridCacheIoManager.access$100(GridCacheIoManager.java:101)
	at org.apache.ignite.internal.processors.cache.GridCacheIoManager$1.onMessage(GridCacheIoManager.java:295)
	at org.apache.ignite.internal.managers.communication.GridIoManager.invokeListener(GridIoManager.java:1569)
	at org.apache.ignite.internal.managers.communication.GridIoManager.processRegularMessage0(GridIoManager.java:1197)
	at org.apache.ignite.internal.managers.communication.GridIoManager.access$4200(GridIoManager.java:127)
	at org.apache.ignite.internal.managers.communication.GridIoManager$9.run(GridIoManager.java:1093)
	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
	at java.lang.Thread.run(Thread.java:748)

As one of the solutions - we shouldn't start such caches on resurrected nodes.
We should save caches changes history somewhere and cluster-widely spread it to joining nodes.
In a case when cache was only stopped, we can do nothing and start it lately when cache start request received.
In a case when cache was stopped & destroyed, we should clean persistence data for that cache.

Issue Links

• causes: IGNITE-12805 Node fails to restart
• causes: IGNITE-12059 DiskPageCompressionConfigValidationTest.testIncorrectStaticCacheConfiguration fails
• causes: IGNITE-12071 Test failures after IGNITE-9562 fix in IGFS suite
• relates to: IGNITE-8717 Move persisted cache configuration to metastore and introduce cache configuration versioning
• links to: GitHub Pull Request #6748
• links to: GitHub Pull Request #6781