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 17:38:35 GMT
I just tried to move view function to separate design doc and no
success - duplicates (with same revision) in view response.


2010/10/7 Paul Davis <paul.joseph.davis@gmail.com>:
> Alexey,
>
> Can you show the other views you have in your design doc? Or
> alternatively, try moving this view to its own design doc?
>
> Paul
>
> On Thu, Oct 7, 2010 at 1:07 PM, Alexey Loshkarev <elf2001@gmail.com> wrote:
>> 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
>>
>



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

Mime
View raw message