-
1. Re: Is that ok to use infinispan Soft Index File Store?
rvansa Sep 10, 2015 4:22 AM (in response to howardpatt)It is experimental until we get enough positive feedback from users trying that out - the 'experimental' tag is just a warning that bugs may show up more often than is usual in the rest of the project. Still, it is a good idea to use it in non-production environment and help us fix bugs similar to the one you're seeing.
So you say that this happens during restart = loading the cache store. Is the data mapped to regular filesystem or is it NFS (NFS may behave strangely). Can you show your configuration? Does that happen if you try to start the server anew with the same data? Is it possible to reproduce with TRACE logging level on org.infinispan.persistence.sifs?
-
2. Re: Is that ok to use infinispan Soft Index File Store?
howardpatt Sep 10, 2015 5:59 AM (in response to rvansa)Hi,Radim:
Thanks for your reply.
1,my file system is ext3
2,yes ,I am trying to restart the server anew with the same data store.
3,maybe I will try TRACE log later
4,my config is like this:
ConfigurationBuilder b = new ConfigurationBuilder(); b.persistence() .passivation(false) .addStore(SoftIndexFileStoreConfigurationBuilder.class) .maxFileSize(1024 * 1024 * 32) .indexSegments(8) .indexLocation(diskStorePath + "/index") .dataLocation(diskStorePath + "/data") .preload(false) .shared(false) .fetchPersistentState(true) .ignoreModifications(false) .purgeOnStartup(false) .async() .enabled(true) .threadPoolSize(10) .eviction() .strategy(EvictionStrategy.LRU) .maxEntries(1000L) .expiration() .lifespan(-1, TimeUnit.SECONDS); cacheManager = new DefaultCacheManager(b.build()); cache = cacheManager.getCache();
any clue?
-
3. Re: Is that ok to use infinispan Soft Index File Store?
howardpatt Sep 10, 2015 7:37 AM (in response to rvansa)this is what the trace log shows:
2015-09-10 19:15:12,716 [TRACE] (sifs.Index$Segment,149) - Indexing IndexRequest{file=16, offset=-7981090, prevFile=-1, prevOffset=-1, size=47, type=UPDATE}
according to what the exception declares,is this negative offset a problem we should look into?
-
4. Re: Is that ok to use infinispan Soft Index File Store?
rvansa Sep 10, 2015 8:07 AM (in response to howardpatt)IndexRequest with negative offset is not a problem - that just denotes that we're indexing that some entry is deleted. The problem is that at some point during the loading either the TemporaryTable or Index itself contains record with negative position.
Do you set expiration explicitly for some of the entries in the store? It seems to me that when startup process runs in parallel with Compactor, it could insert some negative entries. But I don't recall the whole algorithm now, could you upload the whole trace log exhibiting the exception somewhere?
-
5. Re: Is that ok to use infinispan Soft Index File Store?
howardpatt Sep 10, 2015 8:32 AM (in response to rvansa)Unfortunately, the log file is about 17MB , too large to upload, May I email u ?
-
6. Re: Is that ok to use infinispan Soft Index File Store?
rvansa Sep 10, 2015 10:06 AM (in response to howardpatt)I am not sure if my mail will handle 17 MB attachment, could you rather use Dropbox or similar service and post a link here?
-
7. Re: Is that ok to use infinispan Soft Index File Store?
howardpatt Sep 14, 2015 12:30 AM (in response to rvansa)Sorry for the late reply because I had some business to care about during the weekend. Check out this shared link please,you can download the file. THX!
-
8. Re: Is that ok to use infinispan Soft Index File Store?
rvansa Sep 14, 2015 4:27 AM (in response to howardpatt)Hmm, I know what's the problem. I'll try to deal with that this week.
[ISPN-5743] SIFS.start() throws IllegalArgumentException: Negative position - JBoss Issue Tracker
-
9. Re: Is that ok to use infinispan Soft Index File Store?
howardpatt Sep 14, 2015 8:56 AM (in response to rvansa)I really appreciate your reply.
I am looking forward to the fix eagerly, can't wait any longer.
THX again!
-
10. Re: Is that ok to use infinispan Soft Index File Store?
rvansa Sep 18, 2015 4:41 AM (in response to howardpatt)PR ready: https://github.com/infinispan/infinispan/pull/3715
ISPN-5743 was trivial fix, but the fix for ISPN-5753 took me two days to find out.