Log file Viewer
#mongodb logs for Tuesday the 11th of June, 2013
[06:08:17] <shashi> Hey, so I have a basic normalization question. I have Projects having many Tasks and Users, Tasks have many Assignees (who are Users). Now when a user logs in, I want to show all the projects he is part of and all the tasks he is assigned. Also, of course when viewing a task (or project), I show who the assignees are. Should I create a new collection to
[06:08:17] <shashi> store Project - Tasks, Project - Users, Tasks - Assignee relationships? How is this done in day to day practice with mongo?
[07:32:29] <ncopa> is it normal that each /usr/bin/mongo* file is 9.5MB? or have I dont something wrong (link static when it should been dynamic)
[07:45:54] <ProLoser> i only have the db info to the destination, not the source, the source is in mongo_hub, but it hangs when i export
[08:01:01] <ncopa> i bet with some dynamic linking the size of /usr/bin/mongo* could be reduced to 1/10 or so
[08:49:16] <kali> Nevaron_: you need to project/aggregate the other fields you want... look for $sum, $max, $min, etc
[09:35:58] <Nevaron_> Can anyone help me with this please: http://stackoverflow.com/q/17040100/1313143
[09:44:30] <vargadanis> is there maybe a reference DB somewhere that I could use to play around with mondodb?
[09:45:49] <vargadanis> ron, I was thinking more on the lines of a DB export that I could import locally and fiddle around with it
[09:51:43] <vargadanis> Nodex, to be honest, does not really matter... something on which I could experiment with queries, updates etc...
[09:58:02] <vargadanis> why not? :D that's exactly what I was looking for.. some lazy tool that generates data for me hehe
[10:00:59] <stennie> @vargadanis: twitter is a good source of json data which you can import to experiment with. eg: https://search.twitter.com/search.json?q=mongodb
[10:01:58] <stennie> the manual also has example data + queries for aggregation framework: http://docs.mongodb.org/manual/tutorial/aggregation-examples/
[10:17:56] <Derick> you'd want to look at the experimentatl text search in that case (as I suppose you meant with a Regex search there?)
[10:23:15] <shashi> Hello, I have a simple normalization question, I asked before but no one was around to answer I guess.
[10:23:15] <shashi> I am making an app that has Tasks which have many Users (as Assignees). I want to display all the tasks a user is assigned as well as all the users assigned to a task. What is the right way to store data in Mongo for this?
[10:24:21] <ron> it'd be difficult to know which one would suit you best though, without going into quite a bit more details.
[10:24:31] <foofoobar> Derick: no.. Let me example what I want to do: Users should be able to search for a city and all results which are near this city should be showed.
[10:24:55] <ron> but for bidirectional relations between entities, I'd consider either using an external indexing, or using a link entity.
[10:25:24] <foofoobar> 1) Look in mongodb if there is a {shortname: $city}. If not, I make an API call to nominatim (openstreetmap api), get the record and write it into the database
[10:25:45] <foofoobar> the problem is that there are some cities which have the same name, so the "display_name" given from nominati is something like:
[10:26:23] <shashi> ron: I thought of storing assignees inside Tasks, but again, to fetch tasks for a particular user, you'll need to go through all tasks to see if the user is assigned. Right?
[10:26:24] <ron> shashi: nothing to look up really. it's just a collection where you say {source: entity id, target: entity_id} if you need a directed link or just {entities: [id1, id2, id3...]
[10:26:59] <foofoobar> Derick: so I thought of the following: I save two values: ShortName (just "Plittersdorf") and the whole DisplayName (Plittersdorf, Bad ....)
[10:27:03] <ron> shashi: yes, which is why I'd suggest using an external indexer, though in some cases mongo's internal indexing would suffice.
[10:27:19] <ron> again, difficult to say without digging deep into the use case and possible future use cases.
[10:27:22] <foofoobar> Now when a user searches for "Plittersdorf" I will make a lookup for the ShortName. If there are more than one let the user decide
[10:27:34] <shashi> ron: That's what I ended up doing actually. Just wanted to ask you all once if there's anything I could do better
[10:27:40] <foofoobar> is this a good approach or am I over-seeing something which I can do with mongodb?
[10:29:39] <shashi> ron: haha, I understand you'll need to know more about the app to give a more relevant answer. But this is about how much I have thought about the app itself :P
[10:30:41] <shashi> ron: you mean I store {entities: [id1, id2, id3 ...]} both in the tasks and users collection?
[10:31:26] <ron> shashi: the main concern is the query types that you need, possible future queries and which query is done more often. it would also depend on how you update the data. even in a 'small' use case there are many relevant questions.
[10:33:45] <shashi> ron: okay, so there's no usual way to do this. What I did is create a collection that simulates an "edge table" {left_id: source, right_id: target, edgeData: arbitrary_data}
[10:34:21] <ron> shashi: the only thing I can tell you for sure is that you DON"T want to keep a bi-directional link between the entities INSIDE the entities.
[10:35:53] <shashi> Oh yes. When I delete a task, I'll need to make sure all the related entries go away etc :/
[10:36:21] <ron> shashi: yes, that's a way to do it. I'd probably use an external indexing, but that's my preference. I'd normally need to have external indexing for other reasons, so I might as well use it to solve that as well.
[10:36:46] <remonvv> shashi, stuck to MongoDB? Sounds like a graph db or rdbms might be a better fit for you.
[10:37:18] <foofoobar> Derick: okay ;) So I want to realise a "search results which are nearby <city>"
[10:37:37] <shashi> remonvv: I can manage with mongo I think, I am using meteor. So, yes, stuck to mongo
[10:37:51] <foofoobar> Derick: not yet, I have to OSM data set, but this does not contain a simple city+coords
[10:38:30] <foofoobar> so I want to do it like "if entry is stored in db, return this, else sedn request to nominatim-api"
[10:38:55] <foofoobar> because I get values back like "Plittersdorf, Bad Godesberg, Nordrhein-Westphalen, Deutschland"
[10:39:00] <Derick> well, the question is, how do you want to use the data that you store, how are you going to query against it?
[10:41:11] <foofoobar> Derick: The user searches for "plittersdorf" and gets .. lets say 3 results. Now I display the 3 full DisplayNames, so he can take the right one
[10:41:12] <Derick> will you have more than one entry for "plittersdorf" if two users pick a different Long name?
[10:41:54] <remonvv> Provided that's the actual use case that will give you the best performance for the most common tasks.
[10:42:51] <remonvv> On the assumptions that; a) there are significantly less users than tasks eventually, b) you most commonly need query on work for a specific user rather than users for specific work.
[10:42:56] <shashi> remonvv: Let's say it will give good performance for half of the tasks, the other half is showing a list of tasks which the user is assigned
[10:43:58] <remonvv> shashi, that's not a problem. You have two routes for that. Simply execute two queries or, the more common approach in NoSQL, denormalization :
[10:44:25] <foofoobar> Derick: If a user searches for "Plittersdorf" I get multiple results, so I display him both "DisplayNames" and he can choose the right one
[10:44:51] <foofoobar> Derick: If he searches for "Bonn" there is only one result, so he instantly gets this returned
[10:45:12] <foofoobar> Derick: if he searches for "XYZ" which is not in the database, I will make an API call to nominatim and save this into the database
[10:45:21] <shashi> remonvv: hmm, okay what's wrong with storing user _ids instead of the whole user document?
[10:45:51] <remonvv> shashi, there's nothing wrong with it. And I'm certainly not argueing that you embed the user document in the task document.
[10:46:16] <shashi> remonvv: also, I think if I am querying "list of tasks for user X", I'll first need to get all the tasks and go through their assignees to see if X is part of them, no?
[10:46:17] <remonvv> If you just store an ID you're absolutely certain every task->user query will actually require 2 seperate queries.
[10:47:14] <foofoobar> so I will save bandwidth and will not exceed the quota for he nominatim requests
[10:47:14] <remonvv> shashi, I'm telling you to store in the task document what you most typically need for visualisation (username, avatar if supported) but nothing more.
[10:47:23] <Derick> foofoobar: just two tips, use a 2dsphere index for the coordinates and a geojson storage container (isntead of an array with lon, lat)
[10:47:41] <remonvv> That way if you query task X your UI can immediately display assignees. Only when people click on that user will you have to load the profile or something.
[10:47:49] <shashi> remonvv: Ah, okay. Hmm, but I did not really understand why you made the assumption that eventually there will be more tasks than users. Can you explain?
[10:48:30] <remonvv> shashi, most systems that have tasks and users have that pattern. E.g. a project management database would have dozens of users put thousands of tasks.
[10:49:46] <remonvv> shashi, experience. Most systems that manage some sort of tasks have less users than they have tasks.
[10:50:02] <remonvv> tasks are created by users, users are not created by tasks, hence one should generally outnumber the other.
[10:50:17] <shashi> remonvv: I meant why you lead to the conclusion that userids be contained in tasks from there.
[10:52:18] <remonvv> Simple, if a user can have X tasks and a task can have Y users you want to embed task IDs into user if X>Y and vice versa.
[10:53:04] <remonvv> Makes sense right? If you grab a task in this schema you get a task and a few users.
[10:58:00] <foofoobar> Derick: Is it right that as soon as I use a field with the structure { type: <type>, coordinates: <coords } it is interpreted as a geojson field?
[11:05:18] <deepy> is there any good wya I can check the amount of inserts/s I can do on my dev machine?
[11:06:01] <deepy> I think it seems to do at most 2,000 inserts per second, which seems low to me since I'm using setting _id
[11:12:52] <agend> i'm going to keep stats data in mongodb, and the data will be agregated in hrs, days, months
[11:13:43] <agend> what is the best way to get a sum of some value in these different collections in one go?
[11:14:48] <agend> i mean I'd like to ask for sum for hours in some range and then some days in other range, and some months in anoter range
[11:17:01] <kali> it's not what you want to hear, but be aware mongodb is not a good fit for time data
[11:18:25] <Derick> agend: operations in MongoDB *never* happen on more than one collection at a time
[11:19:26] <agend> i'm going to use mongo with python, and was thinking about sending js code to mongo to avoid few db hits
[11:19:31] <Derick> I would suggest that you research whether you can pre-aggregate isntead of storing the full stats
[11:20:07] <agend> i want to pre-aggregate, thats why i will have clicks_hr, clicks_day, clicks_month
[11:21:19] <agend> and when i have to get sum of clicks for something what is not full day/month i have to query few collections
[11:21:39] <Derick> but you haven't shown *what* you are storing, and how... so it's a bit theoretical for now
[11:22:01] <deepy> If I'm just inserting a document with an _id using the java driver, is 5k inserts/s considered good speed? I'm trying it with both my laptops and Windows/Linux gives roughly the same numbers :/
[11:23:13] <deepy> This one is a Lenovo W520 with 8GB of RAM, I was expecting slightly better speed :/
[11:28:14] <deepy> kali: { _id: i } is all I'm inserting, this seems rather small to me and 5k/s seems rather bad becuase of that
[11:29:29] <Derick> deepy: the default is write acknowledged inserts... you might want to turn that off
[11:32:30] <deepy> kali: if it can only handle 5k/s then it's slower than what I wrote myself and I can't replace it
[11:58:52] <stefuNz> Hey, i need to store a change history from my entities (from mysql) and i'm evaluating if mongodb would be suitable. those are literally millions (100 million+) change-entities that i would need to store and i would need to be able to query them by entity_id (the identifier of my identifier), the change type and the change date (like in mysql WHERE entity_id = 123 AND type = 'THATFIELDOVERTHERE' AND when > '14 days ago') - can mongodb
[12:05:57] <stefuNz> kali: it would be possible that the data size would be greater than my memory, how would that behave? i summarized the question in this stackoverflow question: http://stackoverflow.com/questions/17042760/storing-changes-on-entities-in-mysql-scalability -- i'm already using mongodb for some other stuff, but i'm not an expert.
[12:10:54] <kali> stefuNz: it would be slower. depending on what you're using it for, it may still be good enough as long as the index can fit in memory
[12:11:52] <tiriel> what would be the recommended way of rotating mongo journals on a production environment?
[12:12:22] <stefuNz> kali: thank you! do you have another idea how this could be reasonable saved (i mean maybe another database system?)
[12:31:13] <tiriel> oh, neverming, it was an outstanding problem we had and it seems to have gone away now that we updated to the latest version of mongodb
[12:34:57] <zeen> i'm using python/mongokit and a bit confused about a relationship i'm trying to do. i'm very new to nosql way of things so learning. i have World and i have Island, for each island it will belong to a world. I know there are no real joins in mongo. what is the data type for the world_id? should the field be world, or world_id
[12:51:40] <ecornips> Question re replication: I want to replicate a smallish data set (20GB) and keep it up to date across 4 different data centres (UK, US, AU, JP), where each data centre has the entirety of the information. Any of the DCs is able to be Master (low-volume, async updates). Is this possible? The election process looks like it supports Master election quite well. But not sure about the replication aspects across data centres. Is this something that should work ok -
[12:53:47] <ecornips> kali: would there be problems with the higher latency (e.g. 150-200ms between some DCs, others will be < 50ms)?
[12:54:39] <kali> ecornips: other people around here probably have more experience than i have with this
[12:55:53] <ecornips> kali: by chance have you ever done local replication (replica sets I guess), where a node has been down for a couple days, e.g. some hardware failure? I'm quite curious about those sorts of impacts
[12:56:27] <ecornips> I've done a bit of testing locally, and things seem to work quite well, but I'm curious for other peoples experiences
[13:04:32] <starfly> ecornips: I haven't replicated in that scenario with MongoDB, but have with other technologies. If you able to tolerate the latency with writes from an application standpoint, it should work fine with MongoDB
[13:05:18] <ecornips> starfly: writes will be few and far between, so very tolerant there. If you don't mind me asking, what other technologies have you used in similar scenarios?
[13:06:34] <n06> hey do you guys know how much overhead the system profiler adds to I/O operations? Trying to gauge how efficiently im writing to my db
[13:06:43] <starfly> ecornips: yes, both with Oracle Streams and Data Guard, from Europe to US to Japan to Singapore; all worked fine
[13:07:51] <starfly> ecornips: care and feeding from time to time (there's no avoiding WAN issues sometime, even when private)
[13:26:23] <starfly> n06: if you need high write throughput, consider spreading your writes around with sharding
[14:22:55] <hectron> I'm going to be building an application which is going to interact with a thousands of users on a daily basis, and I'm going to require an equal amount of read/writes. I intend to perform on-the-fly analytics on the data collected.
[14:24:36] <hectron> I'm likely going to need to use summary tables. I expect there to be billions of reads/writes in my application daily.
[14:24:59] <starfly> hectron: not sure of the pricing (likely very expensive), but SAP Hana might be an interesting avenue to pursue
[14:27:28] <hectron> I'm just a level 1 developer here, so I was given the task to research viable options. We wanted to leverage Mongo and set up sharding/replicas. However, I do not know much more than what I just stated. We have our own datacenters, so I'm sure that we have the infrastructure.
[14:30:01] <starfly> Cassandra is more of a name-value datastore, not the bells and whistles of MongoDB, but also can scale up much better
[14:31:36] <starfly> Hana is also a column store at the lowest level, but you can reassemble row store representations
[14:31:59] <hectron> Interesting. I will definitely refer to these things when I report to my manager.
[14:32:26] <hectron> I have only gotten this quality of responses from the Drupal community and Linode support.
[14:33:58] <starfly> best of luck, hectron, sounds like you'll have a full plate and an interesting use case
[15:01:44] <agend> is there any way to flatten map-reduce result - i dont want the value sub object ???
[15:06:09] <parox> hey, if i have a mongo collection is there an easy way to get all the unique keys?
[15:10:27] <Nodex> someone has made something on github to get all of your keys but I can't remember what it's called
[15:16:59] <parox> if you mean find() and then parse keys, i think that would be slow with a few thousand docs
[15:18:36] <timstermatic> Hi there. I'm using the AF and I want to group numbers by rounding them up. So 119.10609436035156 would be 119
[15:25:20] <starfly> timstermatic: I'm not familiar with a rounding function, but perhaps you can project result as a string and perform a substring on that to at least truncate to an integer
[15:48:30] <barnes> in the log i get Tue Jun 11 17:15:36.701 ERROR: failed to rename '/var/log/mongo/mongod.log' to '/var/log/mongo/mongod.log.2013-06-11T15-15-36': errno:2 No such file or directory
[15:53:22] <starfly> barnes: you've verified that /var/log/mongo directory exists and the /var/log/mongod.log file exists (admittedly, simple questions)?
[15:54:03] <Derick> Swimming_Bird: as pipeline operators you'd use: group on the key, (sort on the key), limit by x
[15:57:10] <barnes> starfly: yes, I think it is because mongod tries to rename it to /var/log/mongo/mongod.log.2013-06-11T15-15-36 and logrotate to mongod.log.1
[15:57:23] <Swimming_Bird> basically i've got a group members collection, i want to do: where user_id: {"$in": […]}, order_by: 'updated_at': -1, limit: 5, distinct: 'group_id'
[15:58:09] <Swimming_Bird> the problem i'm getting is if i pass multiple user ids who are in the same group, i'll get 2 of the same group back and not get 5 unique groups
[15:59:19] <Swimming_Bird> looks like aggregate supports indexes and limits. so it may be the way to go
[15:59:50] <Derick> you should be able to do : match, group_by: group_id, sort: { updated_at: -1 }, limit 5
[16:11:56] <Swimming_Bird> @Derick: do you know why they have the _id: null in the second example? http://docs.mongodb.org/manual/reference/aggregation/match/#stage._S_match
[16:15:25] <barnes> starfly: https://jira.mongodb.org/browse/SERVER-4905 seems to handle standard logrotate in the future
[18:17:24] <grouch-dev> trying to execute a delete in a replica-set but the delete query always fails, with mongo stacktrace.
[18:19:56] <grouch-dev> during the delete, our secondaries seem to be under extreme load from the replication
[18:20:25] <grouch-dev> we start seeing things like "thinks we are down" on all the servers, then the primary loses majority
[18:20:54] <grouch-dev> and the only operation is the single delete request that was issued from the primary
[18:23:19] <starfly> grouch-dev: approximately how many documents are you attempting to delete? How large is the collection you're deleting from?
[18:26:11] <grouch-dev> if we backup the dataset, and restore onto a single machine, this process works fine.
[18:26:38] <grouch-dev> however, our pre-prod and prod environments consist of 4 mongos and an arbiter
[18:26:47] <starfly> how many distinct status values are there? Is status first in the combined status-date index?
[18:29:15] <grouch-dev> I think there are only 2 states. I am trying to get an answer to the second question...please bear with me
[18:31:40] <starfly> you could have a selectivity issue regarding the index. With that many documents, you'd need to have the date field first in a combined index, otherwise, MongoDB has to effectively scan a very large portion of the index, effectively rendering it suboptimal (putting it mildly)
[18:34:34] <starfly> with indexes (almost regardless of the database technology), you want your most selective field (field with the highest number of distinct values) to be the first field in any combined index, so your query optimizer can work from the most selective to the least selective, which greatly optimizes runtime performance
[18:35:55] <starfly> so, no doubt there's added overhead with replication, but I think you more have an issue with the way that index is defined
[18:36:09] <asciiduck> Currently I'm stowing some log data into a mongo collection so I can use mapReduce on the data. One thing I would like to do is map ip addresses from the log data to the country of origin so I can reduce and find numbers of based on country of origin. I'm trying to do that by making a call out to some of the APIs that offer that information, the problem I am encountering is that the javascript engine that mongo uses for processing map reduce functio
[18:36:49] <grouch-dev> starfly: a more broad question would be "why can't mongo handle a simple delete query"
[18:37:27] <starfly> grouch-dev: you are deleting from a 10 million document collection, if it's not indexed properly, what is simple might take forever to execute
[18:38:06] <grouch-dev> starfly: yes, I could handle that, but it just simply cant handle that command
[18:38:59] <grouch-dev> starfly: I am wondering is there a real reason why (which requires much more education on my part) or is it just a bug?
[18:39:30] <grouch-dev> starfly: I would not expect the database to dump a stacktrace and fail on a delete
[18:40:51] <starfly> grouch-dev: agree that there is likely a bug in the way that the command is being handled, but its performance is entirely predictable if not indexed properly
[18:42:07] <starfly> grouch-dev: anyway, you might want to try revising the index, you can probably see things work as you'd like (and improve other query performance)
[18:55:45] <grouch-dev> starfly: I really appreciate your input starfly. I currently have a task to reimplement the indexing in the application, and that insight will be part of that effort. The application uses spring-mongo-data and @Index attribute on sub-documents, so 90% of our indexing is broken anyhow...
[18:57:56] <leifw> anyone know how to determine whether a client object in the ruby driver is connected to a single mongod or to a replica set?
[19:17:45] <starfly> leifw: you could examine all sessions for all mongod instances (using the REST API or from mongo shell) that are in the replica set in question
[19:18:31] <leifw> I am trying to do this through the ruby driver because I want to skip a ruby driver test when I'm running against a replica set
[19:20:16] <starfly> leifw: so you'd prefer to assume things are available on the back end if you've upped the odds with a replica set?
[19:22:01] <leifw> I'm testing tokumx by running some of the driver tests against it, but there are some changed assumptions so for now I just want to skip the driver tests that I know won't work
[19:23:07] <leifw> I just can't figure out what ruby expression will tell me whether the server that the ruby driver is connected to is standalone or the primary of a replica set
[19:40:27] <starfly> leifw: short of sending an rs.conf() command over the wire and interpreting the results (null if not RS) in Ruby, can't think of anything
[20:28:16] <ikurei> hi. i need to ask a noob question, im new to nosql and mongo and i cant figure this out
[20:29:25] <ikurei> is there any easy way to add a property to a property of a document? like, i have a game with this structure { name : "LOL", platform: "PS5", prices : { "amazon" : 50, "play: 50 } }
[20:30:12] <ikurei> what would be the canonical way to do it? i can do it in a couple of dirty ugly ways, but it's too easy a problem to not have an easy solution
[20:32:20] <starfly> ikurei: you can read the document, update the structure in your code, and update it in MongoDB (using the same _id)
[20:33:06] <ikurei> starfly: that was my only working idea. and that's what i meant, there had to be an easiest way
[20:33:38] <kali> but you'd bette change the structure to prices: [{ store: "amazon", price: 50}, {store: "play", price: 50} ]
[20:33:53] <n06> I have a really strange error popping up as im trying to setup my production cluster. I cant add any shards to the mongos
[20:34:22] <n06> "errmsg" : "exception: SyncClusterConnection::insert prepare failed: 10276 DBClientBase::findN: transport error: <host>:27020 ns: admin.$cmd query: { fsync: 1 } <host>:27020:{}"
[20:34:36] <ikurei> kali: thanks a lot. i tried doing that, but without the quotes aroind prices.steam
[20:35:41] <ikurei> kali: also that was my first design, but i need to do fast accesses at it from the node code. that design i guess will be way better to perform queries
[20:36:19] <kali> ikurei: it's usually better to avoid using "variable stuff" as key name in mongodb
[20:37:09] <kali> ikurei: you're going to need an index for amazon, and index for "play", and add a new index every time you add a store
[20:37:39] <nemothekid> I have a collection with ~20,000,000 documents and I need to update all of them once a day. Its stored on a single database with SSD and enough RAM for the whole dataset. The problem is updates are really slow (so slow that we now batch remove and insert) and while this is happening the database latency for other queries goes through the roof. I thought about using another data store like Redis, but I need to be able to sort these
[20:37:39] <nemothekid> documents for top-N queries. Lastly we thought about inserting to a seperate collections and renaming, but if we eventually decide to shard this collection again, that will be a huge problem. Just looking for advice/similar solutions
[20:38:11] <kali> ikurei: also, it will lead you to concatening strings to craft queries which is a first step towards injection vulnerabilities
[20:41:17] <nemothekid> kali: well I don't really need updates, I just need to replace the data. So if I could have an insert that would kill the previous document instead of dropping the newer one that would be optimal
[20:42:42] <starfly> nemothekid: you could use an active-passive database design, update the passive collection while the active copy takes reads, then swap when the passive collection has been updated. Separate the hosts, storage, etc. to avoid interaction.
[20:42:46] <nemothekid> I thought about that but then I'm left with a collection that I'm adding 20mill documents per day, and the indexes are growing, and the sort performance will start to suffer as well
[20:43:56] <kali> nemothekid: active/passive can be an option, in that case... switch db based on the current day parity...
[20:45:25] <nemothekid> That could work, and it would work if we decided to start sharding again as well. Don't like the changes I would have to make to the application, but it seems like the best option for now. Thanks
[20:47:17] <starfly> nemothekid: we do something similar with a backend for a REST app and it works well, although the setup for swapping is (of course) some work
[22:33:47] <generic_nick> hi, i'm using the pymongo driver and trying to set the read_preference to SECONDARD_PREFERRED
[22:33:54] <generic_nick> but on my unsharded collection mongotop/mongostat are showing no reads whatsoever
[22:52:38] <rpcesar> im getting this weird error in my map reduce after switching from a 2.0.5 single server setup to a 2.4 sharded replica set. Every other process seems to be working as far as I can tell. The error I am seeing is : exception: stale config on lazy receive :: caused by :: $err: "[copilot.SearchPartials] shard version not ok in Client::Context: this shard contains v...", code: 9996
[22:53:20] <rpcesar> im not having much luck finding that error, google returns very limited results, and those seem to be bugs that were closed in other versions
[23:51:16] <dberg> I'm trying to set a cluster locally for testing purposes with a config server, 1 mongos and 3 mongod running as a replica set.
[23:52:08] <dberg> if I log into one of the mongod and run rs.initiate({...}) it seems fine from the logs.