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 17:07:24 GMT
Same problem appears again.
What was done till yesterday:
1. Created new database at node2
2. Replicated from node1 to node2
3. Checked. _all_docs return only unique rows. queue/all returns only
unique rows

After a few hour of stable work, couchdb produce duplicates too.
This time, no duplicate documents (_all_docs has only unique strings),
but duplicate view response.
Remove view index (between couchdb restarts) doesn't help. Couchdb
produce stable duplicates in view.

View function:
function(doc) {
  if (doc.type == "queue") {
    log("BUG TEST id:" + doc._id + ", rev:" + doc._rev);
    emit(doc.ordering, doc);
  }
}

Response:
$ curl http://localhost:5984/exhaust/_design/queues/_view/all
{"total_rows":15,"offset":0,"rows":[
....
{"id":"q_nikolaevka","key":10,"value":{"_id":"q_nikolaevka","_rev":"16181-ae5e5cca96b0491f266bc97c37a88f47","name":"\u041d\u0418\u041a\u041e\u041b\u0410\u0415\u0412\u041a\u0410","default":false,"cars":[],"drivers":[],"ordering":10,"type":"queue"}},
{"id":"q_nikolaevka","key":10,"value":{"_id":"q_nikolaevka","_rev":"16176-3a7bbd128bfb257fd746dfd80769b6fc","name":"\u041d\u0418\u041a\u041e\u041b\u0410\u0415\u0412\u041a\u0410","default":false,"cars":[],"ordering":10,"type":"queue","drivers":[]}},
...
]}


Saw that? Two documents with different revisions in it!

Also, couch.log consists of 3 (!) calls of this function for one document:
[Thu, 07 Oct 2010 16:53:51 GMT] [info] [<0.180.0>] OS Process
#Port<0.2132> Log :: BUG TEST id:q_nikolaevka,
rev:16175-11cedeb529991cf60193d436d1a567e9
[Thu, 07 Oct 2010 16:53:51 GMT] [info] [<0.180.0>] OS Process
#Port<0.2132> Log :: BUG TEST id:q_nikolaevka,
rev:16176-3a7bbd128bfb257fd746dfd80769b6fc
[Thu, 07 Oct 2010 16:53:51 GMT] [info] [<0.180.0>] OS Process
#Port<0.2132> Log :: BUG TEST id:q_nikolaevka,
rev:16181-ae5e5cca96b0491f266bc97c37a88f47


Then I do compact to eliminate old revisions.
And now I have 3 duplicates per q_nikolaevka with same revisions!

I think, I found problem. This document has 1000 revisions in database
and here (http://wiki.apache.org/couchdb/HTTP_database_API) is
described default maximum of 1000 revisions of document.



2010/10/7 Alexey Loshkarev <elf2001@gmail.com>:
> Haha!
> Fresh replication (into new database) eliminates duplicates and I can
> sleep quietly.
>
>
> 2010/10/7 Alexey Loshkarev <elf2001@gmail.com>:
>> P.S. dmesg doesn't show any hardware problems (bad blocks, segfaults
>> and so on).
>> P.P.S. I think, I was migrate 0.10.1 -> 1.0.1 without database
>> replication, so it may be my fault.
>>
>> 2010/10/7 Alexey Loshkarev <elf2001@gmail.com>:
>>> 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
>>>
>>
>>
>>
>> --
>> ----------------
>> 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