Log file Viewer
#mongodb logs for Tuesday the 24th of March, 2015
[01:04:08] <Kota> hey i was in kind of late this morning and no one was around, but i'm having an issue with a duplicate index on a collection. http://puu.sh/gN3tr/7ff013951f.png i've got two _id_ indexes ascending on _id and I can't get rid of either of them
[01:43:15] <Kota> okay that's just messed up... i attempted to make another index, with a different name and different field and it duplicated my _id_ index yet again
[02:06:25] <Kota> { "v" : 1, "key" : { "_id" : 1 }, "name" : "_id_", "ns" : "yhr.chat_pms" } { "v" : 1, "name" : "_id_", "key" : { "_id" : 1 }, "ns" : "yhr.chat_pms" } { "v" : 1, "name" : "_id_", "key" : { "_id" : 1 }, "ns" : "yhr.chat_pms" }
[02:12:06] <Boomtime> Kota: good now please try (copy this exactly): db.system.indexes.find({name:"_id_",ns:"yhr.chat_pms"})
[02:17:13] <Boomtime> would you be amicable in supplying a copy of the bson metadata file after a dump?
[02:27:58] <joannac> Kota: you should get a bson file, that you need to upload somewhere (or I can PM you an email address)
[02:29:30] <Kota> Mon Mar 23 19:21:42.132 DATABASE: yhr to dump/yhr Mon Mar 23 19:21:42.133 yhr.chat_pms to dump/yhr/chat_pms.bson Mon Mar 23 19:21:44.302 2064704 objects Mon Mar 23 19:21:44.302 Metadata for yhr.chat_pms to dump/yhr/chat_pms.metadata.json
[02:48:20] <CyanoIntrant> I'm pretty new to NoSQL, anyone have any comments on how https://www.reddit.com/r/SQL/comments/302g9z/ would work in it?
[04:03:39] <Boomtime> Kota: FYI, all three _id indexes are missing the unique flag - that collection is trouble - you should stop using it asap
[04:09:20] <joannac> Kota: if you could reproduce, that would be great; i haven't been able to reproduce so far
[04:32:42] <Boomtime> if you have a support contract you should have been given a way to download the enterprise version
[04:36:53] <joannac> if you go to http://www.mongodb.org/downloads and pick your distro, it'll tell you whether it does or not
[04:51:42] <Kota> the 3.0 docs say to grant anyAction on anyResource, but i don't know if 2.4 supports grants on stuff
[04:55:16] <joannac> copyTo uses eval. eval requires a bunch of stuff as documented in the second link
[05:41:06] <Kota> i did a test copyTo but stopped it mid copy after about a minute, it had copied half of my collection
[05:41:21] <Kota> i just did a bulk insert with python and the entire database copied in around 30 seconds
[05:42:43] <GothAlice> It would have for short bursts, specifically, if you truly were using bulk operations, at the bulk operation size limit.
[05:45:20] <GothAlice> Using futures, have some easy multiprocess parallelization. What isn't covered by the timings at the top is the resource utilization. Single-process was using ~33% of one core, ~11% on the mongod core. Four-process was using ~80% of each of four cores, 60% of one mongod core. (Much better utilization.)
[05:45:44] <GothAlice> Hmm, that'd be why you're not blocking at all. XD Each operation is quick, and you're not doing any batching at all.
[05:46:23] <GothAlice> (I really ought to be doing proper batching, but, eh… I'm lazy, and this is a rarely-run benchmark I use to test WiredTiger.)
[05:48:34] <Kota> i copied the data to the new collection, renamed the broken index collection, renamed the new collection to what the broken was before, and now the renamed broken collection has two duplicate indexes for _id and the new collection has all of the broken collections old indexes...
[05:48:34] <GothAlice> If the process is aborted part-way through for any reason, though, you'll understandably need to repair the indexes.
[05:49:05] <GothAlice> Sounds like something somewhere is adding indexes automatically. Any ensure_index calls as import-time side effects?
[05:49:42] <GothAlice> Ah, so you weren't spinning up a higher-level application to examine the result, then.
[05:50:52] <Kota> i already sent joannac the bson of my indexes to confirm that they are indeed duplcates...
[05:51:53] <Kota> joannac: none of my copy operations were done through mongo, so its something up with the daemon? or possibly the database itself
[05:52:30] <GothAlice> Kota: Could you repeat the drop/re-copy and only check the result of db.collection.getIndexes() from the mongo shell?
[05:53:16] <GothAlice> (Heh, checking some of mine to make sure duplicate _id_ isn't actually normal, I notice that none of mine are named anything useful: t_p_1_t_o_1__id_1. wut.)
[05:59:28] <Kota> alright. http://hastebin.com/ixiwajahah.json - copied the collection via python. mid-process index snapshot
[06:03:26] <winem_> hi together. I guess some of you will already have passed the m202 course (advanced deployments and operations). can you tell me if the MMS is only part of the stuff for the first week or in all weeks?
[06:04:14] <Kota> i thought this was just a normal okay message, but renaming the new collection did not produce it, but that copy on the broken collection returned this error: http://hastebin.com/ucamevudef
[06:05:25] <GothAlice> Number of records alone isn't sufficient to say, this data is the same as this other data. ;)
[06:06:05] <Kota> i haven't verified every record obviously but at first glance the data is what should be there
[06:06:08] <GothAlice> That type of namespace error is a "bad sign" and warrants more scrutiny, and/or a --repair.
[06:07:47] <winem_> GothAlice, please tell me more about hashing the BSON documents. I would expect issues with timestamps or die id objects. I will wait until Kota is happy and knows his issues fixed :)
[06:08:21] <GothAlice> MongoDB out of the box doesn't validate the wholesale integrity of every BSON value it's handed. (Some things, such as the well-fromedness of records to insert it takes on faith unless an option is given to explicitly check.) This means it's fully possible to do some _really_ crazy things with your BSON documents, either on purpose, or accidentally.
[06:09:13] <GothAlice> For example, one could construct a single document with multiple _id values. (Due to the way the document is stored, repeated keys are technically allowable.)
[06:10:30] <GothAlice> Thought it was 2.6. (They changed the default, apparently in 2.4, to check-by-default.)
[06:10:41] <Kota> well for now i'm going to update the application component that uses that table, since the new collection is fine, only when i rename it to that old name does it screw up
[06:12:48] <Boomtime> https://github.com/mongodb/mongo/blob/r2.4.10/src/mongo/db/namespace_details.cpp#L924
[06:13:30] <Boomtime> somehow, despite the collection not being there, the index assertion of 'not present in destination' is failing.. meaning it trips over something that shouldn't exist
[06:13:58] <Boomtime> when you delete a collection, apparently the _id index is not being deleted with it
[06:14:51] <GothAlice> If there's multiple and an unhanded "multiple documents returned" (or equivalent) exception in the loop (by name?) to clear the indexes…? Fair number of assumption in that hypothesis.
[06:18:25] <GothAlice> Kota: Would now be an opportune time to upgrade to a (supported) version of the DB? :D
[06:18:45] <GothAlice> (I.e. attempt to export, then import, sans indexes, into a newer cluster. Then re-create the "good" indexes.)
[06:18:47] <Boomtime> you are in an impossible state, i would certainly advise dumping out your data
[06:19:37] <GothAlice> Tusker: http://docs.mongodb.org/manual/release-notes/3.0-compatibility/#driver-compatibility-changes
[06:19:52] <GothAlice> Tusker: Also, see the rest of the page, notably the authentication section. (It's all useful to know, though.)
[06:21:36] <GothAlice> (That would have required some rather interesting changes to mongod to achieve. ;)
[06:43:24] <Kota> so, in preparing for mongodump, how should i go about excluding the db indexes? system, System, or system.indexes?
[07:12:18] <girb1> Hi .. is there any way I can restrict IP based access to mongo router without IP tables
[08:02:57] <ZorgHCS> girb1: This is exactly why iptables exist, I don't understand why that's not an option?
[08:06:13] <girb1> ZorgHCS: I need t give only read access for a mongo router , all writes should be forbidden even though I have implemented user auth
[11:52:27] <lxsameer> Is there any work around to store version numbers as keys in the mongodb Hash ?
[11:55:06] <Derick> then again, you are using a *value* as a &key& there - which is not a good thing to do
[11:56:18] <lxsameer> Derick: that won't help my case, which one is faster to query , hashes or embeded documents ?
[12:13:36] <StephenLynx> if you use '$' you are saying that it is a field in the document you are working with
[12:14:09] <Stiffler> hmm.. so how does this line 'tags' : {$pushAll : req.body.tag}, should look like?
[12:15:22] <StephenLynx> http://docs.mongodb.org/manual/reference/operator/update/push/#example-push-each
[13:59:37] <StephenLynx> unlike relational db's, you don't have to keep track of much stuff in mongo.
[14:12:08] <ggoodman> Just had my production 3.0.1 primary crash with an OOM error. Anyone around that can help me diagnose the root cause and help me prevent this from happening again?
[14:35:02] <latestbot> I have referenced an id in an array in another document, how can I remove it in mongodb?
[14:42:37] <latestbot> Suppose I have a field like this in a document, “tags” : [ "550afb5971634aa331d010f3", "550afb6071634aa331d010f4", "550afb6a71634aa331d010f5" ], I just want to remove the first item say. What would I use?
[14:52:20] <Alittlemurkling> Hello #mongodb, I'm attempting to construct a hierarchy of distinct medical license specialties. Doctors can have multiple licenses, so I store them in an array of subdocuments. It basically looks like this: {"_id": "1170", "licenses": [ { "specialtyCategory": "Internal Medicine", "specialtyName": "Sleep Medicine"}, {"specialtyCategory": "Pediatrics", "specialtyName": "Sleep Medicine"}] }
[14:53:40] <Alittlemurkling> I would like something that gives the same result as the distinct command, but have been unable to reproduce it with $aggregate.
[14:55:45] <Alittlemurkling> So far, I have come up with this: db.provider.aggregate({$match: {"licenses.specialtyCategory": "A"}}, {$unwind: "$licenses"}, {$match: {"licenses.specialtyCategory": "A"}}, {$project: {"licenses.specialtyName": 1}})
[14:57:11] <d0x> Any Idea why I get a java.io.IOException: cannot find class com.mongodb.hadoop.mapred.BSONFileInputFormat with Hive on EMR using the Mongo-Hadoop Connector? This Bootstrap script i used: https://s3.eu-central-1.amazonaws.com/christian.emr/bootstrap.sh
[14:58:17] <Alittlemurkling> This is how I tried to use distinct: db.provider.distinct("licenses.specialtyName", {"licenses.specialtyCategory": "Pediatrics"})
[14:59:03] <Alittlemurkling> But that would return both specialtyNames (ah, if they were different. Pardon the simplcity of my example).
[15:15:39] <Alittlemurkling> StephenLynx. Okay. I will just construct the relationship when I import the data.
[15:16:53] <StephenLynx> and these specialties, their hierarchy is completely recursive or do they have a limit?
[15:17:48] <Alittlemurkling> The maximum depth is three levels, but I'm working with a client that only has two currently.
[15:20:40] <StephenLynx> yeah, this is a highly relational model. Mongo will be a very bad tool for this.
[15:21:35] <Alittlemurkling> Nah. This is a cache for a search engine, and I can map this relation when I parse the data.
[15:25:40] <StephenLynx> and this parent specialty is the same as the specialty and could have a parent of its own?
[15:29:43] <Alittlemurkling> There is a license document. It looks like this: { "specialtyCategory": "A", "specialtyName": "B", "subSpecialty": "C", "licenseNumber": "1", "licenseState": "CO" }
[15:32:36] <Alittlemurkling> Given a specialtyCategory, I want all of the distinct specialtyNames associated.
[15:33:28] <StephenLynx> a collection for categories, another one for specialties and another one for sub specialties
[15:33:45] <StephenLynx> a little, yes. but you would have to query only once for what you need anyway.
[15:34:04] <StephenLynx> using fake relations as long as you don't have to perform multiple queries is not an issue.
[15:35:56] <pamp> Hi, I have a compound index {MoName: 1, "P.k": 1}, I can use the index prefix {MoName:1} in a query like db.test.find({MoName:"UtranCell"}).hint({MoName:1})
[15:36:48] <StephenLynx> because then you would have to query for every relation in addition to your original query
[16:39:33] <imachuchu> I have an older mongodb instance (~2.0) and am wondering if there is a way to backup/restore it to a more modern version keeping objectid relationships valid?
[16:41:10] <cheeser> there's no real relationship as such. just common values that your app correlates.
[16:43:11] <imachuchu> cheeser: the dump format changed in 2.2 so importing from an older dump kindof works, but not really. I'm more wondering if there is a conversion script somewhere I'm missing
[16:47:01] <imachuchu> cheeser: IMO that seems a bit ugly, but I don't know what's actually changed internally so I guess that's what I can do
[16:47:40] <StephenLynx> what I would do, install a VM with 3.0, migrate directly and fix whatever broke
[17:58:23] <sellout> cheeser: Sorry, I haven’t done much Java – I assumed I would need to get a reference to the Logger to do that.
[18:02:47] <cheeser> you'd configure in the logging config file for whatever logging framework your app uses.
[18:39:07] <d4rklit3> I have a mongodatabase on compose.io and another one ij ust made on a dedicated server somewhere. Is it possible to clone the one from compose.io to my new one via the mongoshell on my dedicated host?
[20:06:31] <afroradiohead> Hey guys, if I'm reading the document correctly, upsert is a mixture of insert and update right?
[20:13:08] <cheeser> it's better if you use indexes, yes. but it's not predicated upon their existance
[20:15:51] <afroradiohead> I guess I was wondering if I can do an upsert request like "oh these indexed values are duplicates, let me just update this document"
[20:23:28] <StephenLynx> I believe so, but it will count as an index for the document that contains the nested object.
[20:26:41] <afroradiohead> oh definitely not, I've built my own structure. Mapping classes to collections
[20:29:35] <btwotch1> Hi, I am trying to retrieve an object from the mongodb in go via a nested property: http://pastebin.com/tcn4tvtG - does anybody have some helpful comments? thx in advance
[20:37:53] <StephenLynx> latestbot mongo uses BSON that is very similar to JSON. most rules of javascript apply.
[20:39:41] <btwotch1> I am able to retrieve the structure when it is non-nested - I just can't get it to make the query nested
[20:45:39] <btwotch1> ah, you mean: c.Find(bson.M{"tok.token": t}); - but then I cannot even find the query in strace
[20:51:33] <btwotch1> it has no documentation about nested queries: https://www.google.de/?gfe_rd=cr&ei=7c0RVdOGA6mF8QfbkIDICA&gws_rd=ssl#q=site:labix.org%2Fmgo+nest
[20:58:07] <btwotch1> strace helped, I just starred a bit longer and saw, that in the mgo query the '+' is missing "iN+ZWEE"
[21:21:47] <imachuchu> Progz: I personally have no idea, but a bit of googling led to this: http://docs.mongodb.org/v3.0/release-notes/3.0-upgrade/
[21:28:34] <Progz> I thought mongoDB put index into ram. I have one db with 23 indexes for a index size of 8 GB. But My ram consumed is 500 MB.
[21:34:32] <imachuchu> Progz: no, I think it's only best if the indexes fit into ram. If the db hasn't been queried for a while/at all since starting the server it's completely normal to have them not loaded yet
[21:35:38] <Progz> I have more than 1000 websites online. and I have a lot of request in my db. but nothing in ram
[21:38:31] <GothAlice> Progz: You might want to look at your "page fault" counts. If the number is high (which it very likely will be) then MongoDB can't efficiently utilize RAM because the kernel is constantly swapping out chunks behind-the-scenes. (This makes things very, very slow.)
[21:39:43] <GothAlice> Note that the majority of the memory MongoDB uses isn't allocated, it's memory-mapped on-disk files. These will typically show up not as "real" usage, but as "caches", because the kernel can freely move things around to free up the RAM if something else needs it.
[21:45:43] <Progz> ok GothAlice so when I see with mongostat in the column faults number around 132. is it a lot ?
[21:54:02] <GothAlice> (res size would be the "working set" memory used to process queries, map/reduce, and manage connections, etc.)
[21:59:26] <GothAlice> Well, at these scales, it doesn't really matter how much RAM you have, it won't be enough. ;)
[22:00:36] <GothAlice> 30GB of RAM, assuming 10% overhead in general, means you have 14% as much RAM as you need for that dataset to fit entirely. What's the aggregate index size? (The size of all indexes across all collections?) Not sure the easiest way to get this information other than by getting db.stats() on each DB.
[22:02:22] <GothAlice> Progz: Well then. The page faults are the kernel going out to disk to answer a query by MongoDB. This means it'll suffer amazingly bad performance, since you're going back to disk for potentially large chunks of data, in some cases, more than 200 times per second. You really, really, really want to invest in sharding at this point.
[22:03:19] <GothAlice> Notably also http://docs.mongodb.org/manual/tutorial/deploy-shard-cluster/ and http://docs.mongodb.org/manual/tutorial/choose-a-shard-key/ to ensure your data gets "evenly" spread.
[22:04:11] <GothAlice> This is MongoDB's official way to split data between multiple hosts in order to bring the per-host data set size below the RAM size of the individual hosts and regain optimal performance.
[22:04:32] <d4rklit3> nvm i got it was in the conf file, however I can't seem to be able to connect to the server with auth.
[22:04:45] <Progz> ok GothAlice I am following a mooc in the mongodb university. I was thinking about sharding but never try do configure one ^^
[22:06:00] <GothAlice> d4rklit3: Several potential issues: did you upgrade to 3.0 from any prior version? If your DB host is 3.0, are your client drivers up-to-date and compatible? (Both ref: http://docs.mongodb.org/manual/release-notes/3.0-compatibility/) Have you already added users? If not, you'll need to use the localhost exception to add the first user, which should be an admin. For this, ref: http://docs.mongodb.org/manual/tutorial/enable-authentication/
[22:06:32] <Progz> I misunderstanding mongo documentation, I was thinking we can only put index in ram not all data.
[22:06:37] <GothAlice> d4rklit3: Please review the links I have provided to ensure you aren't running into compatibility issues.
[22:07:29] <GothAlice> Progz: Alas, that doesn't quite work. Think of it like a "most recently used" cache of file pages in RAM. Some of those pages will be indexes, sure, but to the kernel, it has no way other than frequency of access to determine if a page is "hot or not", so in some cases your indexes might get paged out to make room for other data being paged in.
[22:09:07] <GothAlice> d4rklit3: The other hint of mine was going to be "are you authenticating against the right database?" Since MongoDB users are local to a database, such as the "admin" DB, you may have to specify if you created them in the wrong one.
[22:10:51] <GothAlice> In the last few days I've seen it doing terrible, terrible things, like creating duplicate indexes on _id, failing 3.0-style authentication, etc.
[22:12:19] <GothAlice> Seriously, what does everyone find so attractive about robomongo over a syntax highlighting shell? I find the "(0) {…}" list items to be purposefully obstructing, not helpful. :/ It hides information from you, and doesn't empower you in terms of querying any more than a bare shell. :/
[22:13:00] <Progz> GothAlice: I am not an expert of mongo so robomongo help me find right command :)
[22:13:22] <GothAlice> Stop trying to use ancient (no longer supported, two major versions behind) tools against the latest and greatest server.
[22:13:58] <GothAlice> If you want authentication in 3.0, you need _all_ of your tools to be 3.0-aware and 3.0-compatible.
[22:14:50] <d4rklit3> so server 1 is node app, server 2 is mongo. server 1 app needs mongo. it uses some adaptor for it
[22:15:41] <GothAlice> Progz: In terms of finding commands, nothing beats the documentation, which also explains the output of those commands. ;) Even I refer back to the documentation a lot. (There's a reason I'm so quick on the docs.mongodb.org links…)
[22:16:17] <GothAlice> d4rklit3: Indeed. AFIK it hasn't been updated to the 3.0 authentication scheme yet.
[22:17:05] <GothAlice> However, I'll repeat my warning from earlier: I've seen robomongo do *destructively* terrible things to people's data in the last week.
[22:18:31] <GothAlice> That's one of the issues encountered: just connecting and fetching a collection created a duplicate _id index for one user. (There may, or may not have been other things going on, but _not_ connecting robomongo after cloning the collection from one to another in Python didn't exhibit the duplicate index, firing robomongo up, it spontaneously appeared.)
[22:19:24] <d4rklit3> the IT guys instaleld 3.0 on this server and left me with the config!! im just a programmer!
[22:19:29] <GothAlice> At least mongoose issues have historically been "it just does things weirdly and isn't particularly well documented when they choose to be wacky".
[22:20:06] <GothAlice> d4rklit3: After the 16th user storing ObjectIds as strings came through the channel, mongoose got added to my blacklist. ;)
[22:20:43] <GothAlice> Additionally, failure of documentation, which at several points effectively suggests a broken approach.
[22:21:38] <GothAlice> Indeed. Efficient, meaningful 12-byte storage of creation timestamp, application server ID, process ID, and auto-increment (per server process) counter.
[22:22:34] <GothAlice> Turn it into a string (as many mongoose users do), or worse, mix real ObjectIds and strings, and the support incidents begin. ;)
[22:32:12] <GothAlice> d4rklit3: Yeah, pardon my Python-esque call syntax in the term. Pretend those are "{_id: " instead of "_id=". ;)
[22:58:04] <ThisIsDog> I'm using PyMongo 2.8 in my application. When I add a document to the database and it already exists, a DuplicateKeyError exception is thrown. When I do e.details on the exception, I'm always getting None. Under what circumstances is details not None?
[22:58:58] <ThisIsDog> I'm trying to get the document existing in the database that caused the DuplicateKeyError
[23:14:48] <ehershey> you can tell by running in a shell: db.SomeRandomMadeUpNewCollection.insert({}); db.SomeRandomMadeUpNewCollection.getIndexes()