[
https://issues.apache.org/jira/browse/JCS-73?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13493065#comment-13493065
]
Thomas Vandahl commented on JCS-73:
-----------------------------------
Ok, I did some experiments.
- If you add the thread name to the key, the problem disappears. That means
that the actual problem is the concurrent writing of the *same* key to the
cache. The write is not synchronized and as such not atomic. That means that
writing the same object into the cache concurrently is bound to fail. See
GroupCacheAccess.java:166.
- The problem is not specific to the IndexedDiskCache but also appears (even
more drastically) when the auxiliary is not configured which proves the
assumption above.
So we are making progress...
> concurrent cache access causes values loss
> ------------------------------------------
>
> Key: JCS-73
> URL: https://issues.apache.org/jira/browse/JCS-73
> Project: Commons JCS
> Issue Type: Bug
> Components: Indexed Disk Cache
> Affects Versions: jcs-1.3
> Reporter: Alexander Kleymenov
> Assignee: Thomas Vandahl
> Fix For: jcs-2.0.0
>
> Attachments: patch.txt, repro.zip
>
>
> The following groovy code demonstrates the problem: several threads write to
> the cache and try to read previously written values. Sometimes just written
> values can't be retrieved. In this case, subsequent tries can get the value,
> but not always.
> {code}
> cache = org.apache.jcs.JCS.getInstance("cache")
> group = "group"
> worker = {
> def name = Thread.currentThread().name
> 10000.times { idx ->
> if (idx) {
> // get previously stored value
> def res = cache.getFromGroup(idx-1, group)
> if (!res) {
> // null value got inspite of the fact it was placed in cache!
> println "ERROR: for ${idx} in " + name
> // try to get the value again:
> def n = 5
> while (n-- > 0) {
> res = cache.getFromGroup(idx-1, group)
> if (res) {
> // the value finally appeared in cache
> println "ERROR FIXED for ${idx}: ${res} " + name
> break
> }
> println "ERROR STILL PERSISTS for ${idx} in " + name
> Thread.sleep(1000)
> }
> }
> }
> // put value in the cache
> cache.putInGroup(idx, group, [value:[a:1, b:2, c:3], aux:[1:'a',
> 2:'b', 3:'c', t:name]])
> if (!(idx % 10000)) {
> println name+" "+idx
> }
> }
> }
> Thread.start worker
> Thread.start worker
> Thread.start worker
> Thread.start worker
> Thread.start worker
> Thread.start worker
> Thread.start worker
> Thread.start worker
> {code}
> Cache configuration:
> {code}
> jcs.default=CACHE
> jcs.default.cacheattributes=org.apache.jcs.engine.CompositeCacheAttributes
> jcs.default.cacheattributes.MaxObjects=-1
> jcs.default.cacheattributes.MemoryCacheName=org.apache.jcs.engine.memory.lru.LRUMemoryCache
> jcs.default.cacheattributes.DiskUsagePatternName=UPDATE
> jcs.default.cacheattributes.UseMemoryShrinker=true
> jcs.default.cacheattributes.MaxMemoryIdleTimeSeconds=10
> jcs.default.cacheattributes.ShrinkerIntervalSeconds=10
> jcs.default.elementattributes=org.apache.jcs.engine.ElementAttributes
> jcs.default.elementattributes.IsRemote=false
> jcs.default.elementattributes.IsLateral=false
> jcs.default.elementattributes.IsSpool=true
> jcs.default.elementattributes.IsEternal=true
> jcs.auxiliary.CACHE=org.apache.jcs.auxiliary.disk.indexed.IndexedDiskCacheFactory
> jcs.auxiliary.CACHE.attributes=org.apache.jcs.auxiliary.disk.indexed.IndexedDiskCacheAttributes
> jcs.auxiliary.CACHE.attributes.DiskPath=./cache
> jcs.auxiliary.CACHE.attributes.MaxPurgatorySize=-1
> jcs.auxiliary.CACHE.attributes.MaxKeySize=-1
> jcs.auxiliary.CACHE.attributes.MaxRecycleBinSize=500
> jcs.auxiliary.CACHE.attributes.ShutdownSpoolTimeLimit=60
> jcs.auxiliary.CACHE.attributes.OptimizeAtRemoveCount=30000
> jcs.auxiliary.CACHE.attributes.OptimizeOnShutdown=true
> jcs.auxiliary.CACHE.attributes.EventQueueType=SINGLE
> {code}
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
