couchdb-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alexey Loshkarev <elf2...@gmail.com>
Subject Re: view response with duplicate id's
Date Thu, 07 Oct 2010 09:27:16 GMT
I think, this is database file corruption. Query _all_docs returns me
a lot of duplicates (about 3.000 duplicates in ~350.000-documents
database).


[12:17:48 root@node2 (~)]# curl
http://localhost:5984/exhaust/_all_docs > all_docs
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100 37.7M    0 37.7M    0     0  1210k      0 --:--:--  0:00:31 --:--:--  943k
[12:18:23 root@node2 (~)]# wc -l all_docs
325102 all_docs
[12:18:27 root@node2 (~)]# uniq all_docs |wc -l
322924


Node1 has duplicates too, but very small amount:
[12:18:48 root@node1 (~)]# curl
http://localhost:5984/exhaust/_all_docs > all_docs
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100 38.6M    0 38.6M    0     0   693k      0 --:--:--  0:00:57 --:--:-- 55809
[12:19:57 root@node1 (~)]# wc -l all_docs
332714 all_docs
[12:20:54 root@node1 (~)]# uniq all_docs |wc -l
332523



2010/10/7 Alexey Loshkarev <elf2001@gmail.com>:
> I can't say what specific it may be, so let dive into history of this
> database(s).
>
> First (before a 5-6 weeks) it was node2 server with couchdb v10.1.
> There was testing database on it. There were alot of structural
> changes, view updates and so on.
> Than it becomes production and starts working ok.
> Than we realize we need backup, and best - online backup (as we have
> couchdb we can do this).
> So, there appears node1 server with couchdb 1.0.1. I replicated node2
> to node1, than initiates continuous replication node1 -> node2 and
> node2 -> node1. All clients works with node2 only. All works fine
> about a month.
> Few days before we was at peak load, so I'v want to use node1 and
> node2 simultaneously. This was done by round-robin on DNS (host db
> returns 2 different IP - node1's ip and node2's IP). All works fine
> about 5 minutes, than I gave first conflict (view queues/all returns
> two identical documents, one - actual version, second - conflicted
> revision, document with field _conflict="....."). Document ID was
> q_tsentr.
> As I don't has conflict resolver yet, I resolves conflict manually by
> deleting conflicted revision. I'v also disables round-robin and move
> all load to node2 to avoid conflicts for a while to wrote conflict
> resolver.
>
> It works ok (node1 and node2 in mutual replications, active load on
> node2) till yesterday.
> Yesterday operator call me he has duplicate data in program. At this
> queues/all returns 1 duplicated document - the same as few days before
> (id = q_tsentr). One row consists of actual document version, another
> row consists of old revision with field _conflicted_revision="some old
> revision".
>
> I tried to delete this revision but without success. GET for
> q_tsentr?rev="some old revision" returns valid document. DELETE
> q_tsentr?rev="some old revision" gaves me 409 error.
> Here are log files (node2):
>
> [Wed, 06 Oct 2010 12:17:19 GMT] [info] [<0.7239.1462>] 10.0.0.41 - -
> 'GET' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229 200
> [Wed, 06 Oct 2010 12:17:30 GMT] [info] [<0.7245.1462>] 10.0.0.41 - -
> 'GET' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229 200
> [Wed, 06 Oct 2010 12:17:35 GMT] [info] [<0.7287.1462>] 10.0.0.41 - -
> 'GET' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229 200
> [Wed, 06 Oct 2010 12:17:43 GMT] [info] [<0.7345.1462>] 10.0.0.41 - -
> 'GET' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229 200
> [Wed, 06 Oct 2010 12:18:02 GMT] [info] [<0.7864.1462>] 10.0.0.41 - -
> 'DELETE' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229
> 409
> [Wed, 06 Oct 2010 12:18:29 GMT] [info] [<0.8331.1462>] 10.0.0.41 - -
> 'GET' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229 200
> [Wed, 06 Oct 2010 12:18:39 GMT] [info] [<0.8363.1462>] 10.0.0.41 - -
> 'DELETE' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229
> 409
> [Wed, 06 Oct 2010 12:38:19 GMT] [info] [<0.16765.1462>] 10.0.0.41 - -
> 'GET' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229 200
> [Wed, 06 Oct 2010 12:40:40 GMT] [info] [<0.17337.1462>] 10.0.0.41 - -
> 'GET' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229 200
> [Wed, 06 Oct 2010 12:40:45 GMT] [info] [<0.17344.1462>] 10.0.0.41 - -
> 'DELETE' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229
> 404
>
> Logs at node1:
>
> [Wed, 06 Oct 2010 12:17:46 GMT] [info] [<0.25979.462>] 10.20.20.13 - -
> 'GET' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229 200
> [Wed, 06 Oct 2010 12:17:56 GMT] [info] [<0.26002.462>] 10.20.20.13 - -
> 'DELETE' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229
> 200
> [Wed, 06 Oct 2010 12:21:25 GMT] [info] [<0.27133.462>] 10.20.20.13 - -
> 'DELETE' /exhaust/q_tsentr?rev=all 404
> [Wed, 06 Oct 2010 12:21:49 GMT] [info] [<0.27179.462>] 10.20.20.13 - -
> 'DELETE' /exhaust/q_tsentr?revs=true 404
> [Wed, 06 Oct 2010 12:24:41 GMT] [info] [<0.28959.462>] 10.20.20.13 - -
> 'DELETE' /exhaust/q_tsentr?revs=true 404
> [Wed, 06 Oct 2010 12:38:07 GMT] [info] [<0.10362.463>] 10.20.20.13 - -
> 'GET' /exhaust/q_tsentr?revs=all 404
> [Wed, 06 Oct 2010 12:38:23 GMT] [info] [<0.10534.463>] 10.20.20.13 - -
> 'GET' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229 200
> [Wed, 06 Oct 2010 12:40:25 GMT] [info] [<0.12014.463>] 10.20.20.13 - -
> 'GET' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229 200
> [Wed, 06 Oct 2010 12:40:33 GMT] [info] [<0.12109.463>] 10.20.20.13 - -
> 'DELETE' /exhaust/q_tsentr?rev=27144-f516ac68e697874eef9c7562f3e2e229
> 404
>
> So, I deletes this document and creates new one (id - q_tsentr2).
> It will works fine about hour.
>
> Node2 has undeletable duplicate, so I move all clients to node1. There
> were now such problem, view response was correct.
>
> Than I tried to recover database at node2. I stops, deletes view index
> files and start couchdb again. Than i ping all view to recreate index.
> At the end ot this procedure, i saw duplicates of identical rows (see
> first letter in this thread). Node1 has no such problems, so I stops
> replication, leave load on node1 and go for crying into this maillist.
>
>
> 2010/10/6 Paul Davis <paul.joseph.davis@gmail.com>:
>> It was noted on IRC that I should give a bit more explanation.
>>
>> With the information that you've provided there are two possible
>> explanations. Either your client code is not doing what you expect or
>> you've triggered a really crazy bug in the view indexer that caused it
>> to reindex a database without invalidating a view and not removing
>> keys for docs when it reindexed.
>>
>> Given that no one has reported anything remotely like this and I can't
>> immediately see a code path that would violate so many behaviours in
>> the view updater, I'm leaning towards this being an issue in the
>> client code.
>>
>> If there was something specific that changed since the view worked,
>> that might illuminate what could cause this sort of behaviour if it is
>> indeed a bug in CouchDB.
>>
>> HTH,
>> Paul Davis
>>
>> On Wed, Oct 6, 2010 at 12:24 PM, Alexey Loshkarev <elf2001@gmail.com> wrote:
>>> I have such view function (map only, without reduce)
>>>
>>> function(doc) {
>>>  if (doc.type == "queue") {
>>>    emit(doc.ordering, doc.drivers);
>>>  }
>>> }
>>>
>>> It works perfect till yesterday, but today it start return duplicates
>>> Example:
>>> $ curl http://node2:5984/exhaust/_design/queues/_view/all
>>>
>>> {"total_rows":46,"offset":0,"rows":[
>>> {"id":"q_mashinyi-v-gorode","key":0,"value":["d_mironets_ivan","d_smertin_ivan","d_kasyanenko_sergej","d_chabotar_aleksandr","d_martyinenko_yurij","d_krikunenko_aleksandr"]},
>>> {"id":"q_mashinyi-v-gorode","key":0,"value":["d_mironets_ivan","d_smertin_ivan","d_kasyanenko_sergej","d_chabotar_aleksandr","d_martyinenko_yurij","d_krikunenko_aleksandr"]},
>>> {"id":"q_mashinyi-v-gorode","key":0,"value":["d_mironets_ivan","d_smertin_ivan","d_kasyanenko_sergej","d_chabotar_aleksandr","d_martyinenko_yurij","d_krikunenko_aleksandr"]},
>>> ......
>>> {"id":"q_oblasnaya","key":2,"value":["d_kramarenko_viktor","d_skorodzievskij_eduard"]},
>>> {"id":"q_oblasnaya","key":2,"value":["d_kramarenko_viktor","d_skorodzievskij_eduard"]},
>>> {"id":"q_oblasnaya","key":2,"value":["d_kramarenko_viktor","d_skorodzievskij_eduard"]},
>>> ........
>>> {"id":"q_otstoj","key":11,"value":["d_gavrilenko_aleksandr","d_klishnev_sergej"]}
>>> ]}
>>>
>>>
>>> I tried to restart server, recreate view (remove view index file),
>>> compact view and database and none of this helps, it still returns
>>> duplicates.
>>> What happens? How to avoid it in the future?
>>>
>>> --
>>> ----------------
>>> Best regards
>>> Alexey Loshkarev
>>> mailto:elf2001@gmail.com
>>>
>>
>
>
>
> --
> ----------------
> Best regards
> Alexey Loshkarev
> mailto:elf2001@gmail.com
>



-- 
----------------
Best regards
Alexey Loshkarev
mailto:elf2001@gmail.com

Mime
View raw message