[jira] [Created] (IGNITE-5564) Race between read-through and topology version update

["Alexey Goncharuk (JIRA)" <ji...@apache.org>]

Alexey Goncharuk created IGNITE-5564:
----------------------------------------

             Summary: Race between read-through and topology version update
                 Key: IGNITE-5564
                 URL: https://issues.apache.org/jira/browse/IGNITE-5564
             Project: Ignite
          Issue Type: Bug
          Components: cache
    Affects Versions: 1.7
            Reporter: Alexey Goncharuk
            Assignee: Alexey Goncharuk
             Fix For: 2.1


I occasionally observe the following assertions when working with ATOMIC cache with cache store on changing topology:
{code}
java.lang.AssertionError: Invalid version for inner update [isNew=false, entry=GridDhtAtomicCacheEntry [super=GridDhtCacheEntry [rdrs=[], locPart=GridDhtLocalPartition [id=157, map=org.apache.ignite.internal.processors.cache.GridCacheConcurrentMapImpl@7a99d0af, rmvQueue=GridCircularBuffer [sizeMask=31, idxGen=0], cntr=8, shouldBeRenting=false, state=OWNING, reservations=0, empty=false, createTime=06/21/2017 09:59:03], super=GridDistributedCacheEntry [super=GridCacheMapEntry [key=KeyCacheObjectImpl [val=1181, hasValBytes=true], val=CacheObjectImpl [val=1181, hasValBytes=true], startVer=1498028394357, ver=GridCacheVersion [topVer=109508344, time=1498028344708, order=1498028394358, nodeOrder=1], hash=1181, extras=GridCacheTtlEntryExtras [ttl=60000, expireTime=1498028404707], flags=0]]]], newVer=GridCacheVersion [topVer=109508343, time=1498028344709, order=1498028394369, nodeOrder=1]]
	at org.apache.ignite.internal.processors.cache.GridCacheMapEntry.innerUpdate(GridCacheMapEntry.java:2311)
	at org.apache.ignite.internal.processors.cache.distributed.dht.atomic.GridDhtAtomicCache.updateSingle(GridDhtAtomicCache.java:2485)
	at org.apache.ignite.internal.processors.cache.distributed.dht.atomic.GridDhtAtomicCache.updateAllAsyncInternal0(GridDhtAtomicCache.java:1887)
	at org.apache.ignite.internal.processors.cache.distributed.dht.atomic.GridDhtAtomicCache.updateAllAsyncInternal(GridDhtAtomicCache.java:1727)
	at org.apache.ignite.internal.processors.cache.distributed.dht.atomic.GridNearAtomicAbstractUpdateFuture.mapSingle(GridNearAtomicAbstractUpdateFuture.java:264)
	at org.apache.ignite.internal.processors.cache.distributed.dht.atomic.GridNearAtomicSingleUpdateFuture.map(GridNearAtomicSingleUpdateFuture.java:494)
	at org.apache.ignite.internal.processors.cache.distributed.dht.atomic.GridNearAtomicSingleUpdateFuture.mapOnTopology(GridNearAtomicSingleUpdateFuture.java:436)
	at org.apache.ignite.internal.processors.cache.distributed.dht.atomic.GridNearAtomicAbstractUpdateFuture.map(GridNearAtomicAbstractUpdateFuture.java:209)
	at org.apache.ignite.internal.processors.cache.distributed.dht.atomic.GridDhtAtomicCache.update0(GridDhtAtomicCache.java:1242)
	at org.apache.ignite.internal.processors.cache.distributed.dht.atomic.GridDhtAtomicCache.put0(GridDhtAtomicCache.java:675)
	at org.apache.ignite.internal.processors.cache.GridCacheAdapter.put(GridCacheAdapter.java:2294)
	at org.apache.ignite.internal.processors.cache.distributed.near.GridNearAtomicCache.put(GridNearAtomicCache.java:437)
	at org.apache.ignite.internal.processors.cache.GridCacheAdapter.put(GridCacheAdapter.java:2271)
	at org.apache.ignite.internal.processors.cache.IgniteCacheProxy.put(IgniteCacheProxy.java:1379)
{code}

The assertion happens because there is a race between these events:
1) An update is mapped on topology version N
2) Topology version changes and discovery updates the version to N+1, but the event is not yet processed by the exchange future
3) A read-through request comes in and performs the read. Inside {{versionedValue()}} call a new entry version is generated. Since the discovery version is already updated, the new entry version is based on topVer=N+1
4) Update request proceeds and read-locks the topology. Since the exchange future is not yet initialized, the request does not attempt to remap and proceeds with version N
5) The next entry version is generated using request topology version = N
6) Inside the entry update method, we assert that new version is greater than old version, but it's not the case in this scenario, and assertion fails

Attached is a test reproducing the issue (see testConcurrentReadThroughUpdate())



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)