invoking flush while indexing and rebalancing causes issues ( MB4594 still a problem in dp4. Undefined set view `default` for `_designXXX` after rebalance )

Filipe Manana added a comment - 11/Apr/12 6:50 AM Farshid, can your team try to reproduce this? This might be a consequence of issue MB-5046 (which Trevor just created).

Filipe Manana added a comment - 11/Apr/12 6:56 AM Added ns_server component, as it might be related. However this is likely a consequence of MB-5046 .

Aleksey Kondratenko added a comment - 11/Apr/12 10:34 AM We need logs from problematic nodes. Give as output of /diag. This can be done through browser by clicking 'Generate diagnostic report' on Logs page

Trevor Murphy added a comment - 11/Apr/12 12:30 PM Attached some log files from relevant node. Diagnostic report too big

Aleksey Kondratenko added a comment - 11/Apr/12 1:05 PM I'll take a look at what you gave, but how about posting diag compressed?

Aleksey Kondratenko added a comment - 11/Apr/12 1:56 PM Information you posted is useful but in order to understand this issue completely and fix it for DP5 I need full diag.

Peter Wansch added a comment - 05/Jul/12 10:36 AM Farshid, can your team reproduce if this still exists and close otherwise?

Farshid Ghods added a comment - 05/Jul/12 10:57 AM this happened because user called flush on the database. we will try and reproduce this with flush after flush_all is implemented or fixed for 2.0

Peter Wansch added a comment - 19/Jul/12 12:22 PM Farshid, can you close this as obsolete because this feature is no longer supported?

Farshid Ghods added a comment - 19/Jul/12 12:49 PM flush_all is still supported and can be enabled via flushctl

Dipti Borkar added a comment - 03/Aug/12 7:12 PM Is there a way to get a better error message from view engine when flush zaps the index files. Also refer to MB-5170 .

Filipe Manana added a comment - 05/Aug/12 1:25 PM There's nothing I can do here. What happens is apparently that the bucket is deleted, and sooner or later the index files are deleted from disk (as it should be), and then recreated but never configured (or maybe some delay) by ns_server on at least one of the nodes. Maybe some race condition somewhere due the async vbucket deletes, but the message is clear about what the problem is. Also, not sure if this is still reproducible with latest builds, DP4 codebase is too old now.

Filipe Manana added a comment - 06/Aug/12 4:30 AM Farshid, can someone from your team do the tests with latest builds?

Farshid Ghods added a comment - 06/Aug/12 10:13 AM I did run a test and tried flush_all when indexing and things failed. the bug is also closed as dupe because flush_all is not implemented yet and we are waiting for flush_all to work on 2.0 I am going to close this as dupe since ALL flush_all bugs are also closed as dupe regardless of their end result and error we are seeing.

Farshid Ghods added a comment - 06/Aug/12 10:14 AM flush_all does not work on 2.0 builds. once flush_all is fixed we will run more tests and reopen the bug if it is still applicable

Dipti Borkar added a comment - 06/Aug/12 10:23 AM Farshid, This bug is specifically about the error message from the view engine. Can we leave it open until we test flush_all? (See my comments above)

Farshid Ghods added a comment - 06/Aug/12 10:31 AM sounds good. QE will test this once flush_all is actually fixed.