[system test] failed to delete data, cannot locate database file

Thuan Nguyen added a comment - 10/Oct/12 9:47 PM Integrated in github-ep-engine-2-0 #439 (See [ http://qa.hq.northscale.net/job/github-ep-engine-2-0/439/ ])      MB-6844 requeue delete operation until the snapshot task is complete (Revision 03360987688eb408bdecefe844b843fb16702e04) MB-6844 Increase max retry number for couchstore_open_db (Revision 4f9646ca6ade4fc072165eb97571daf0bde6b9df)      Result = SUCCESS Jin Lim : Files : * src/ep.cc Jin Lim : Files : * src/couch-kvstore/couch-kvstore.hh * src/couch-kvstore/couch-kvstore.cc

Jin Lim added a comment - 10/Oct/12 1:18 PM QE tested the toy build with the pending patch set below. No commit failure for delete was found. http://review.couchbase.org/#/c/21463/

Jin Lim added a comment - 10/Oct/12 1:13 PM Thanks for the comment. Unfortunately it isn't really easy to distinguish btw persisting write commit failure and temp. write commit failure. EP Engine made a modification that increase the number of retry for commit failure that eventually reduce the change of spitting out the false(temp) write commit failure message all over the place. http://review.couchbase.org/#/c/21454/

Farshid Ghods added a comment - 09/Oct/12 2:57 PM Jin, we need to make that clear for the user so that they dont sense a major issue when they see this error. does the current log message in the UI mention which one of these cases we are hitting ? if thats the case then we should have this in the 2.0 release notes documentation

Jin Lim added a comment - 09/Oct/12 1:08 PM There are two different warnings due to similar Couchstore file i/o access error: 1) Write commit failure due to open_db error, no such file 2) Delete data failure due to readdir error, no file is found for the given name The first case (1) is a normal behavior under the current design of how the compactor deleting a old db file and renaming a newly compacted db file with a new revision number. There is always a narrow window where CouchKVStore cannot open the old db file while the new db file is being compacted. The retry logic in EP engine takes care of this window. According to the log, the failed writes eventually got succeeded after being re-queued by the retry logic thus no persisting write failures on the given vbuckets. The second case is a real issue we need to continue to look at. Please either open a new bug (and close this bug) with the description of (2) or update the current bug accordingly. Thanks~

Jin Lim added a comment - 08/Oct/12 10:30 PM OK, thanks.

Thuan Nguyen added a comment - 08/Oct/12 9:49 PM Rebalance was able to complete without any issue. Cluster continued running load with mutation, deletion and expiration. So I could not say if it had any data loss.

Jin Lim added a comment - 08/Oct/12 9:36 PM Thanks Tony. So this is not the delete noop case since each file has doc count > 0. However, I don't still see the failures are persisting after retry (we always have a window of retry during the compaction). Anyway, will take a look at the full logs. Were there any data loss or rebalance failure due to the write commit failure?

Jin Lim added a comment - 08/Oct/12 9:36 PM Thanks Tony. So this is not the delete noop case since each file has doc count > 0. However, I don't still see the failures are persisting after retry (we always have a window of retry during the compaction). Anyway, will take a look at the full logs. Were there any data loss or rebalance failure due to the write commit failure?

Chiyoung Seo added a comment - 08/Oct/12 9:14 PM Jin, Tony just uploaded the log files and db_info output.

Thuan Nguyen added a comment - 08/Oct/12 9:12 PM - edited Here are the output from Jin command [ root@ip-10-248-109-239 tmp]# /opt/couchbase/bin/couch_dbinfo /data/saslbucket/499.couch.1 DB Info (/data/saslbucket/499.couch.1)    file format version: 10    update_seq: 8803    doc count: 8743    deleted doc count: 60    data size: 8.63 MB    B-tree size: 755.0 kB    total disk size: 9.45 MB [ root@ip-10-248-109-239 tmp]# /opt/couchbase/bin/couch_dbinfo /data/saslbucket/122.couch.1 DB Info (/data/saslbucket/122.couch.1)    file format version: 10    update_seq: 8821    doc count: 8753    deleted doc count: 68    data size: 8.60 MB    B-tree size: 756.8 kB    total disk size: 9.42 MB * Disk write commit also failed in other node /opt/couchbase/var/lib/couchbase/logs/memcached.log.1.txt:Mon Oct 8 07:35:11.881001 UTC 3: Warning: failed to load the engine session stats due to IO exception "basic_ios::clear" /opt/couchbase/var/lib/couchbase/logs/memcached.log.2.txt:Mon Oct 8 18:27:57.567943 UTC 3: Warning: failed to load the engine session stats due to IO exception "basic_ios::clear" /opt/couchbase/var/lib/couchbase/logs/memcached.log.3.txt:Mon Oct 8 18:29:46.512624 UTC 3: Warning: failed to parse the session stats json doc!!! /opt/couchbase/var/lib/couchbase/logs/memcached.log.4.txt:Mon Oct 8 21:24:24.698305 UTC 3: Warning: failed to load the engine session stats due to IO exception "basic_ios::clear" /opt/couchbase/var/lib/couchbase/logs/memcached.log.4.txt:Tue Oct 9 00:26:30.516362 UTC 3: Warning: couchstore_open_db failed, name=/data/saslbucket/355.couch.1 option=0 rev=1 retried=2 error=no such file [none] /opt/couchbase/var/lib/couchbase/logs/memcached.log.4.txt:Tue Oct 9 00:26:30.516422 UTC 3: Warning: failed to open database, vbucketId = 355 fileRev = 1 numDocs = 32 /opt/couchbase/var/lib/couchbase/logs/memcached.log.4.txt:Tue Oct 9 00:26:30.516433 UTC 3: Warning: commit failed, cannot save CouchDB docs for vbucket = 355 rev = 1 /opt/couchbase/var/lib/couchbase/logs/memcached.log.4.txt:Tue Oct 9 00:28:48.725180 UTC 3: Warning: failed to delete data, cannot locate database file /data/saslbucket/552.couch.1 /opt/couchbase/var/lib/couchbase/logs/memcached.log.4.txt:Tue Oct 9 00:28:48.725562 UTC 3: Warning: failed to delete data, cannot locate database file /data/saslbucket/552.couch.1 /opt/couchbase/var/lib/couchbase/logs/memcached.log.4.txt:Tue Oct 9 00:28:48.725970 UTC 3: Warning: failed to delete data, cannot locate database file /data/saslbucket/552.couch.1 /opt/couchbase/var/lib/couchbase/logs/memcached.log.4.txt:Tue Oct 9 00:28:48.726365 UTC 3: Warning: failed to delete data, cannot locate database file /data/saslbucket/552.couch.1 [ root@ip-10-248-109-80 tmp]# /opt/couchbase/bin/couch_dbinfo /data/saslbucket/355.couch.1 DB Info (/data/saslbucket/355.couch.1)    file format version: 10    update_seq: 8911    doc count: 8601    deleted doc count: 62    data size: 8.50 MB    B-tree size: 747.1 kB    total disk size: 9.63 MB [ root@ip-10-248-109-80 tmp]# /opt/couchbase/bin/couch_dbinfo /data/saslbucket/552.couch.1 DB Info (/data/saslbucket/552.couch.1)    file format version: 10    update_seq: 8860    doc count: 8714    deleted doc count: 73    data size: 8.59 MB    B-tree size: 772.2 kB    total disk size: 9.77 MB

Thuan Nguyen added a comment - 08/Oct/12 9:08 PM Link to collect info of all nodes https://s3.amazonaws.com/packages.couchbase/collect_info/ec2/20121008/8nodes-1808-ec2-colinfo-write-commit-failed-20121008-180102.tgz

Jin Lim added a comment - 08/Oct/12 9:08 PM Please see my comments requesting further information.

Jin Lim added a comment - 08/Oct/12 9:07 PM If the delete data was being issued after either flush/delete vbucket 499, it will be a noop. The logging behavior has slightly changed but the noop logic has been there since the early inception of 2.0. Also after flushing and deleting a given vbucket, ep engine recreates a empty vbucket (db file) with the same vbucket id. Thus I bet 499.couch.1 & 122.couch.1 are empty in the above case. If the system is still alive, please run the following and paste the result here as well. ~/install/bin/couch_dbinfo ~/data/saslbucket/499.couch.1 (and 122.couch.1) Thanks!

Jin Lim added a comment - 08/Oct/12 9:03 PM Looking at the logs above, those warnings look harmless. Can't really tell without the full log but all looks apparently the case where subsequent retry must have succeeded. Please attache the full logs here. Thanks.

Chiyoung Seo added a comment - 08/Oct/12 8:54 PM Jin, We still saw the commit failure due to "file not found". Please take a look at it to see if there are any regressions from our recent changes in couchkvstore.

Karan Kumar added a comment - 08/Oct/12 8:42 PM Chiyoung, Let me know if this does not fall into ep_engine project. I will re-assign it accordingly.