Can't reproduce it with this script. If I send you copy of my buggy database, will it help to improve couchdb?
2010/10/7 Paul Davis <[email protected]>: > Alexey, > > I tried writing a script that'd hammer a single document and make > random view requests to see if I could reproduce the issue. Currently > the test doc is on revision 6,000 or so and I've not reproduced your > issue. I've included the script below, can you try it or see anything > that I should add to try and get closer to your situation? > > Paul > > #! /usr/bin/env python > > import random > import couchdbkit > > def main(): > server = couchdbkit.Server("http://127.0.0.1:5984/";) > db = server.get_or_create_db("foo") > > ddocid = "_design/baz" > if ddocid not in db: > db[ddocid] = {"views": {"foo": {"map": """ > function(doc) {emit(doc._id, doc.value);} > """}}} > > assert 0 <= len(db.view("baz/foo")) <= 1 > > docid = "foo" > > if docid in db: > doc = db[docid] > else: > doc = {"_id": "foo", "value": 1} > db[docid] = doc > > for i in range(1200): > db[docid] = doc > doc = db[docid] > if random.random() < 0.33: > assert len(db.view("baz/foo")) == 1 > > if __name__ == '__main__': > main() > > > On Thu, Oct 7, 2010 at 1:46 PM, Alexey Loshkarev <[email protected]> wrote: >> If it helps.. >> >> This q_* documents are some such of state data. They are changed very >> frequently. >> I have 12 q_* documents and they may be changed 10-30 time per minute. >> May be, there are race condition problem in couchdb in view creation? >> >> >> 2010/10/7 Alexey Loshkarev <[email protected]>: >>> 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 <[email protected]>: >>>> 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 <[email protected]> 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 <[email protected]>: >>>>>> Haha! >>>>>> Fresh replication (into new database) eliminates duplicates and I can >>>>>> sleep quietly. >>>>>> >>>>>> >>>>>> 2010/10/7 Alexey Loshkarev <[email protected]>: >>>>>>> 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 <[email protected]>: >>>>>>>> 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 r...@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 r...@node2 (~)]# wc -l all_docs >>>>>>>> 325102 all_docs >>>>>>>> [12:18:27 r...@node2 (~)]# uniq all_docs |wc -l >>>>>>>> 322924 >>>>>>>> >>>>>>>> >>>>>>>> Node1 has duplicates too, but very small amount: >>>>>>>> [12:18:48 r...@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 r...@node1 (~)]# wc -l all_docs >>>>>>>> 332714 all_docs >>>>>>>> [12:20:54 r...@node1 (~)]# uniq all_docs |wc -l >>>>>>>> 332523 >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> 2010/10/7 Alexey Loshkarev <[email protected]>: >>>>>>>>> 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 <[email protected]>: >>>>>>>>>> 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 >>>>>>>>>> <[email protected]> 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:[email protected] >>>>>>>>>>> >>>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> -- >>>>>>>>> ---------------- >>>>>>>>> Best regards >>>>>>>>> Alexey Loshkarev >>>>>>>>> mailto:[email protected] >>>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> -- >>>>>>>> ---------------- >>>>>>>> Best regards >>>>>>>> Alexey Loshkarev >>>>>>>> mailto:[email protected] >>>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> -- >>>>>>> ---------------- >>>>>>> Best regards >>>>>>> Alexey Loshkarev >>>>>>> mailto:[email protected] >>>>>>> >>>>>> >>>>>> >>>>>> >>>>>> -- >>>>>> ---------------- >>>>>> Best regards >>>>>> Alexey Loshkarev >>>>>> mailto:[email protected] >>>>>> >>>>> >>>>> >>>>> >>>>> -- >>>>> ---------------- >>>>> Best regards >>>>> Alexey Loshkarev >>>>> mailto:[email protected] >>>>> >>>> >>> >>> >>> >>> -- >>> ---------------- >>> Best regards >>> Alexey Loshkarev >>> mailto:[email protected] >>> >> >> >> >> -- >> ---------------- >> Best regards >> Alexey Loshkarev >> mailto:[email protected] >> > -- ---------------- Best regards Alexey Loshkarev mailto:[email protected]
