Log file Viewer
#mongodb logs for Wednesday the 3rd of October, 2012
[02:05:14] <ketema> looking for help and discussion on a question I posted: http://kjkh.me/SBsIQd Thanks!
[02:53:29] <jwilliams> we get the following errors Assertion: 13282:Couldn't load a valid config for after 3 attempts. Checking user group shows that upgrading to 1.8 (this seems an 1.6 bug) would fix the problem. but we use 2.0.1.
[05:00:17] <meson10> How does connection Pooling work.. Is it suggested to create a new connection per request or should it be cached at module level and reUsed ?
[05:11:15] <andoriyu> (not just update all documents with one function, but update each document separetly with different values)
[05:12:57] <meson10> andoriyu: No. You can look at the aggregation framework, or write server side JS code.
[05:28:53] <whatasunnyday> Hi! I was reading the mongo manual and I have a simple question. If I make an index that is unique and I insert a document that has that field repeated, will it not be inserted into the collection? Will it return an error?
[05:29:18] <whatasunnyday> Or will it appear stored in the collection but not accessible unless I remove the index?
[06:12:54] <whatasunnyday> http://www.mongodb.org/display/DOCS/Choosing+a+Shard+Key i was thinking that time would be an example of bad sharding but it doesn't seem to be the case. if you shard by one attribute and that attribute grows beyond 64 mb, does it mean it doesn't create another shrad?
[06:19:59] <michaeltwofish> I've just upgraded a server to use MongoDB 2.2.0 from 1.8.3. Starting the shell shows the correct version but db.serverStatus() still shows 1.8.3. Is this cause for concern and something is screwed up?
[06:49:48] <iksik> hmm, still testing and learning about replSets... yesterday, when i was removing master nodes one by one, 'primary' node was selected after less then 1 sec(?) today i'm looking on replset status table and hmm http://i.imm.io/Gw65.png - no primary node o.O
[06:52:28] <oskie> maybe there isn't consensus to elect a node as primary. the log should probably tell
[06:53:42] <oskie> i'm guessing you'll need 4 nodes in a 6 node replica set in order to get consensus
[06:55:15] <iksik> so how can i deal with it when random number of nodes goes down? (for examples, an OS crash)
[06:56:19] <oskie> for instance, a 5 node replica set would survive intact as long as 3 or more nodes are up
[06:56:58] <oskie> personally i don't see why you'd want to have anything other than 3 node replica sets
[07:03:12] <iksik> oskie: one more question... how many resources can be used by an arbiter instance? cpu/mem/bandwidth?
[07:15:00] <ezakimak> question: it says in the manual under Dot Notation vs Subobjects that key order must be the same, and that ... "This can make subobject matching unwieldy in languages whose default document representation is unordered.". However, in the manual page for updating in the "Field (re)order" section it says: "There is no guarantee that the field order will be consistent, or the same, after an update"
[07:16:34] <ezakimak> how can it expect the client to test w/the "correct" key order if the server won't guarantee that it will preserve the same order it was originally written with?
[08:09:49] <arussel> I have a field with a set in it. What is the query to check if an element is part of it ?
[08:16:05] <NodeX> arussel : if you have an array then you can just do foo:"bar" and it will match ["bar","baz","joe"]
[08:18:47] <NodeX> for example if you have [1,2] and you want only docs with 1,2 then it would be foo:[1,2]
[08:35:20] <Lujeni> Hello - someone can explain what does this error mean pls ? http://sebsauvage.net/paste/?fd747367994a0cd3#xwA7ebeaAIPPg2cOmCeAz4b2lAgBH1jt9y6LbARX1Mg=
[08:38:38] <kali> Lujeni: it just means you have an index that has empty physical pages. it happens all the time in collection with heavy write/update/delete rate
[10:46:01] <WormDrink> Hi, I'm using mongodb 1.8 - with following config (http://pastebin.com/SasVcUHK) - I then stop all replica set memebers except the ones with id 3 (status afterward can be seen here: http://pastebin.com/ys4pKnWJ) - then at this point in time I get error whenever I try connect to router : uncaught exception: error { "$err" : "socket exception", "code" : 11002 }
[10:48:10] <WormDrink> I see there is some info regarding this in echo 'db.shards.find()' | mongo localhost:27219/config (localhost:27219is config db server)
[13:29:02] <imbusy> I've set up two load balanced apache servers as workers(16 threads per process, ~25 processes) to run mod_wsgi + mongoengine + replica set with two instances. I'm stress testing the setup and the logs of the primary mongodb instance say that there are ~530 connections open. I'm calling connection.end_request() every time the rendering of the page is done. After a while, if no requests to the servers are made, the database stops responding, but t
[13:29:02] <imbusy> here still remain ~530 open connections. When i restart the apache instances, the connections drop and everything is fine again.
[13:29:11] <imbusy> Question is: is the connection timing out somewhere? why does the database stop responding after not making any requests for a while?
[14:00:23] <ekristen> I'm planning on two replicasets, my understanding for data to be sync'd between the two replicasets is to have have each collection shard'd is that correct?
[14:04:44] <ekristen> NodeX: I'm looking at http://www.mongodb.org/display/DOCS/Simple+Initial+Sharding+Architecture
[14:07:46] <NodeX> you have to decide if you need to or want to shard and also whether high availability in your case is for your entire database or if you're okay with parts of it being available
[14:08:54] <ekristen> NodeX: ok, so let me make sure I understand, the shard is the horizontal in that link, IE between the replica sets?
[14:10:44] <ekristen> http://www.mongodb.org/display/DOCS/Sharding+Introduction <-- architecture overview
[14:11:51] <NodeX> if you can live with parts of your data being offline then shard + replicaset is the highest availability
[14:12:16] <NodeX> because it ensures at least X percent is online (ruling out power failure in 3 data centers)
[14:13:32] <ekristen> NodeX: can I have a replicate set with multiple nodes, spread across different data centers to retain 100% uptime?
[14:14:51] <ekristen> if I shard an entire database across multiple replica sets doesn't that mean id have 100% up time?
[14:15:55] <ekristen> NodeX: correct, I'm just trying to make sure I understand how it works properly
[14:18:23] <NodeX> if you have 3 replicas of your data and the master goes down then an election is held and one of the replicas takes the master's place
[14:20:11] <NodeX> if you need either 100% or 0% then 9 replicasets is better than 3 shards + 3 Replicas for each shard
[14:21:23] <NodeX> imo, if you're unlucky enough to have 3 or more RS fail then you might aswel call it a day!
[14:21:35] <ekristen> NodeX: sure, makes sense, if there are certain collections that are only needed for lets say administrative use vs regular use
[14:22:01] <NodeX> you do pay a price for Replicasets (RS) though, they'll need to be the same size boxes to the master
[14:24:05] <NodeX> I can't remember if you can force a read from a RS or not, prolly best to ask someone who does RS alot
[14:26:12] <ElGrotto> nodex: I think you can, but I'm a noobie who only heard about the existance of nosql yesterday lol.. referencing http://docs.mongodb.org/manual/core/replication/ tho in the Consistency section)
[14:28:42] <imbusy> i'm using pymongo+apache mod_wsgi. I open a connection and it stays open. I call end_request() when I finish processing. If the connection stays idle for about 3 minutes, I can not send any more requests through it, the database simply does not respond. Does anyone have a clue what is going on?
[14:33:04] <ElGrotto> I have what's prob a noob question tho; is it possible to have more than one server accepting writes in a multiple server setup? Assume I can ensure in software that they don't step on each other's toes (say, software instance 1 will only write to rows where some column A=1, instance 2 where A=2, etc).. I'm happy with reads being "eventually consistent".
[14:35:13] <ElGrotto> ah see I think I may have joined just a minute too late in the conversation above :) hehe
[14:42:14] <_m> Reference: http://api.mongodb.org/python/current/faq.html#how-do-i-change-the-timeout-value-for-cursors
[14:42:39] <ekristen> if you have 3 shards and you shard a collection, that means that collection is chunked across all three shards?
[14:43:52] <imbusy> i'm playing with connectTimeoutMS setting it to 30000 right now, but it's not helping - the connections stay open way past the 30s limit
[14:44:35] <imbusy> 5 minutes in, still 614 connections open (~800 were open during peak time) and the database is not responding to any more calls
[14:53:44] <imbusy> i don't think i completely understand how connection pooling works accross threads. where is the pool stored?
[15:14:20] <ElGrotto> kali: ty, and sorry for the delayed response, I was re-reading about sharding.. my understanding is that for that to work, A must be a primary key (it isn't - I want all A=1's to go to one server regardless of their _id).. and I want to choose which server.. let me give a more concrete example.. if A is the user's location, and 1 is France say, and 2 is USA, I would want all A=1 writes to go to the server located in france. I also do want to read
[15:14:20] <ElGrotto> from any server, but I'm happy with the A != 1 servers being a little behind. My understanding is that sharding works by splitting the primary key into chunks and distributing those chunks, which is not the same :S
[15:17:39] <ElGrotto> .. also the french servers would connect to the french instance of mongodb and only write to rows where A=1. Maybe I'm just getting myself confused here :S
[15:21:21] <kali> ElGrotto: ok, several things here. i thnk you can do the chunking yourself (you need to check that)
[15:22:27] <kali> ElGrotto: and there are some new features that I only know from the changelog because they're useless to me at the current time around rack, tagging and datacenter awareness, so maybe you can use these to get what you're looking for
[15:33:16] <aster1sk> Greetings all, attempting to use the lovely new aggregate framework to sum some page view counts. Only problem is I wont have a list of the page id's to group. Here's an example :
[15:34:56] <ekristen> at what point is it smart to shard? I know thats a very subjective question, but I am trying to get an understanding when it is overkill
[15:35:31] <ElGrotto> from the manual, shard when you anticipate more data than your server can handle (in ram or on disk)..
[15:36:12] <ekristen> ElGrotto: so, if I anticipate 1mil documents in a collection with a total size of 4gb and I only have 4gb of ram, I should shard?
[15:40:29] <zastern> I had an unclean mongodb shutdown - I'm trying to get it started with mongod repair, but I'm getting things like exception: "dbpath (/data/db/) does not exist". I'm pretty sure that's not where my database is stored . . . but there's nothing in the man page to tell me how to specify a config, location, etc. Mongo 1.8.x
[15:41:08] <aster1sk> The Aggregate Framework documentation is unclear whether it is possible to sum subdocuments of unknown id's
[15:41:22] <ElGrotto> ekristen: http://docs.mongodb.org/manual/core/sharding/ section "Indications" -- sorry the last page wasn't quite the right one :S
[15:43:19] <zastern> but i have no idea how to tell mongo to use that config when i start it from the command line, because the man pages are basically empty
[15:44:38] <zastern> logs now say remove log file and run repair. which is weird, because this page - http://www.mongodb.org/display/DOCS/Durability+and+Repair - says DONT remove the lock file
[15:45:17] <zastern> NodeX: yes but the docs arent talking about when its running. theyre talking about when it wont start due to unclean shutdown
[15:55:11] <ekristen> do you have define collections to shard, or if you just define sharding for a database will all collections shard?
[15:56:33] <syskk> this works fine when i have a user document whose passports array contains an identical object
[15:59:15] <aster1sk> Using the aggregate framework is it possible to (I'm using PHP) match : _id : array(1,2,3,4) etc? OR must I append and $and?
[16:02:25] <ekristen> can I enable sharing but only have 1 shard to start and then add additional shards as I need them or is it easier to just enable sharding and scale when the time comes to use it?
[16:05:40] <aster1sk> aggregate : { pages : { 1234 : { views : 1 }, 4321 : { views : 5 } } } <--- Is it even possible to sum views in this scenario?
[16:18:13] <syskk> I have this object: { "_id" : ObjectId("506c5c39733b7e89e13a9222"), "name" : "Test", "passports" : [ { "a" : "b", "c" : "d" } ] }
[16:20:30] <imbusy> why does the database stop responding if I keep the connection open for a few minutes? I'm using pymongo
[16:23:25] <ekristen> do I need to do anything special with my rails application to work with mongodb that has implemented shards?
[16:35:37] <imbusy> could that be an issue with mongo or is my cloud provider killing connections that don't send anything over for a few minutes?
[16:39:45] <ag4ve> can i input data and have mongo report on the records that deviate from a schema?
[16:43:02] <kali> ag4ve: seriously no. the whole point of the document model is too free you from the schema.
[16:43:26] <ag4ve> what i want to do is grab generally standard json data and shove it into a store. however, in order to be able to write better queries, i'd like some type of report that a different schema was used
[16:44:25] <ekristen> how much is the config server used in a shard setup, can it be small server or does it need to be as large as or the same size as the nodes being used in a shard cluster?
[16:45:03] <ag4ve> oh, so the schema that mongoose uses is purely sugar and has nothing to do with mongo... i know i read that, i guess i just didn't get it
[16:47:05] <kali> ag4ve: yep, the closest thing mongodb has to a schema is the list of index and their properties
[16:47:26] <ekristen> kali: I am planning the infrastructure for a large application (50k users in first year, 200k second year or more) -- can I stand up one replicate set as a single shard, with a single config server, so that later when I need to expand I just need to add additional shards, or do I need to leave shard disabled until I am ready to have more than one shard?
[16:48:44] <ekristen> kali: so I could have 3 config servers, 1 shard (consisting of 3 nodes in 1 replica set) to start, then expanding later would be easier yes?
[16:48:46] <kali> erkules|away: as for the sharding, it is quite relatively easy to add it beforehand, or to start with a single shard
[16:51:43] <kali> ekristen: make it clear with your dev that you plan to do that though. and think about what your sharding key will be. sharding break a few features that are easy to avoid ( group() opertors )
[16:51:46] <aster1sk> I need to query multiple values on the same key find( { i : [ 1,2,3,4,5 ] } ) -- amidoingitright?
[16:56:23] <aster1sk> local GNU screen + ssh irssi screen... get's a little wacky with the shortcuts.
[16:57:05] <aster1sk> Didn't want to flood other channel with /join -window #mongodb (you guys deserve the utmost attention) therefore I'm using pidgin for this window... too many things.
[16:59:01] <ElGrotto> ok I've been reading about shards for a looong while now and the one thing I've discovered for certain.. it gives me a headache :S
[17:00:19] <kali> ElGrotto: if i may... don't overthink it right now. your job right now is to get the right features for your 50k. goind from 50 to 200k is next year problem
[17:01:18] <ElGrotto> I'm trying to figure out whether it's possible to have a database spread over geographic locations..
[17:02:19] <ElGrotto> remembering I hadn't ever heard about mongo/nosql until yesterday, my mind has gone racing over the possibilities..
[17:03:06] <aster1sk> I've spent nearly 1.5 months just reading the docs / practising... still feel total n00b
[17:03:10] <ElGrotto> wondering whether it could achieve stuff I'd found impossible with sql, because of its lack of tables
[17:03:48] <kali> ekristen: there is a compromise to find. in my current job, we have worried about scalabitly much too early, and implementing anything became complex much too early
[17:04:10] <kali> ekristen: so there is a sweet spot to find. starting with mongodb is a bette start than we had though :)
[17:04:46] <aster1sk> Our CTO really wants read preference so M/R is out of the question (can't run command on master) so I'm hustlin trying to figure out aggregation framework for as many queries as I can.
[17:05:39] <aster1sk> Oh totally, however the framework wasn't available during initial R&D - so I had to change a lot of stuff along the way (esp 2.1 release).
[17:06:33] <ElGrotto> I do disagree with one point though, your *design* has to be scaleable even if your current *implementation* isn't scaled.
[17:07:28] <ElGrotto> that sounds like hacking a solution.. making "improvements" as you find out you're reaching a limit
[17:07:30] <aster1sk> Funny when experimenting with our data model I cooked up a few hairbrained solutions to deal with not having to use that silly positional operator... turns out now I'm stuck in a bit of a quandary.
[17:09:03] <kali> ElGrotto: it was not a comment to be taken literraly. i'm just commenting on a real life experience and advising against premature optimization and overdesign
[17:09:08] <aster1sk> Came back to bite me now though, I can't $sum a $group because there's no finite key of what to $group... so I'll either have to array_sum() it in the model or keep digging... I'm running out of steam.
[17:15:12] <ElGrotto> leads to a bit of a question tho.. the integer equivalent in mongo (I've not seen anything on datatypes if they even exist), is it 32 or 64 bit?
[17:15:51] <aster1sk> Derick would you know how to $group $sum this : aggregate : { pages : { 1234 : { views : 1 }, 4321 : { views : 3 } } } ?
[17:18:19] <aster1sk> Well instead of upserting on the fly, I'd have to find it first then insert / update -- which means two queries.
[17:20:36] <aster1sk> I'm thinking it might be time to separate this schema into different collections as much as I don't want to.
[17:21:12] <Derick> aster1sk: yeah, keep one for the view separately? Might be better as it would avoid moving things around on disk if the document keeps growing
[17:22:05] <Derick> http://www.mongodb.org/display/DOCS/Updating#Updating-The%24positionaloperator has a similar example actually :)
[17:22:12] <aster1sk> I want to be as forthcoming as possible with the data model (it's far more complicated than that) but boss would get mad if I share the damn thing.
[17:22:49] <aster1sk> I also have have subdocuments under pages describing unique views... it gets pretty scary.
[17:28:22] <FraFraFra> morning guys! I'm looking for a good strategy for finding intersection between two huge collections. Huge = ~2M of docs. Collection 1 contains all ids, Collection 2 contains let say 200k missing ids from Collection 1. I would find these 200k
[17:36:34] <kali> FraFraFra: you'll have to do that out of mongo. i would export the ids from both collection with mongoexport, then sort and join them in shell
[17:36:41] <crudson> FraFraFra: if ids are unique within each collection, perform two map reduces, the second of which uses {:out => {:reduce => 'collection'}}. Map key will be the id, and value will be a count.
[17:44:42] <FraFraFra> crudson: ids are the same, I'm using the collection numeber 2 just a temporary collection where I stored just the complete list of Ids cames from Collection 1, minus the missing ids
[19:05:52] <jmpf> I keep getting this - http://pastie.org/private/yuc0bphpqk3342qudhjuaw on 2.2.0 - but both the server and the one backing it up have tons of space - any other gotchas to be aware of?
[19:53:28] <airportyh> Hello all, is it possible in the aggregation framework to group by values in an array field?
[20:01:56] <ezakimak> question: it says in the manual under Dot Notation vs Subobjects that key order must be the same, and that ... "This can make subobject matching unwieldy in languages whose default document representation is unordered.". However, in the manual page for updating in the "Field (re)order" section it says: "There is no guarantee that the field order will be consistent, or the same, after an update"
[20:01:57] <bhosie> i have an admin user that i log in using this: use admin -> db.auth({'user':'pass'}) after setting up a replica set, i can no longer log in. are auth and the keyfile required by replication incompatible?
[20:02:00] <ezakimak> how can it expect the client to test w/the "correct" key order if the server won't guarantee that it will preserve the same order it was originally written with?
[20:35:42] <airportyh> Hello all, I am writing a batch job that uses the aggregation framework to summarize some data, but after getting the response from aggregate, what is a good way to delete all the data I have processed so far?
[20:46:09] <airportyh> http://stackoverflow.com/questions/12716599/mongodb-batch-jobs-and-deleting-data
[20:49:09] <leehambley> should the server delete the database and index files when I drop the database ?
[20:49:40] <leehambley> I'm on a MBA, and the 20Gb of files that mongo creates when my tests run are sortof out of control, I expected deleting the database would take care of those