Log file Viewer
#mongodb logs for Thursday the 11th of February, 2016
[02:00:58] <nyexpress> I am dealing with quite a dilemma here: I have a user who has money in their account, it fluctuates as deposits and withdraws are made. I would like to calculate the total transacted amount over the period of a month, and it's becoming quite a challenge to figure out how to approach it, my equation is as such: ( $today['withdraws'] - $lastMonth['withdraws'] ) + ( $today['deposits'] - $lastMonth['deposits'] ), however a user can clear their
[09:43:10] <Ange7> someone can help me and can explain me how to do this : http://pastebin.com/Xk3bnhYY
[09:49:31] <Ceunincksken> Is my understanding correct that you have a collection that contains 3 fields (_id, user id and a date) ?
[09:51:41] <Ceunincksken> You enter a Date range and you want all the users that lies within this range?
[09:52:43] <Ange7> I enter a date range and i want all documents that appear only once in my collection within this range.
[09:58:31] <Lope> I want to run something like FindAndModify, but I want it to modify and return multiple documents. But the documentation says: "The findAndModify command modifies and returns a single document."
[10:00:30] <Lope> Okay I'll just do a normal find on all the docs, then update them in another command.
[10:00:58] <Ceunincksken> @Ange7: Can you read the following PasteBin and tell me if this is what you want? http://pastebin.com/AaACip8g
[10:22:25] <Ceunincksken> You can combine all operators in one pipeline such as group, match, geonear, limit, ...
[10:45:56] <Keksike> I need to somehow monitor what kind of queries go into my mongodb and get some kind of stats from them (whats their operation-time, do they hit indexes, etc.). What monitoring tool can I use to log this information?
[10:48:51] <newbsduser> hello, i have very big dataset, if i don't have enough memory, mongodb responses go very slowly, can ssd disks with raid can solve the io performance problem? (for instance 6 x 450G with raid10)
[11:00:06] <Ceunincksken> Keksike: You have 2 options. Run your query with explain() or set the profiling level on your database.
[11:01:29] <Ceunincksken> newbsduser: Going with SSD disks can solve your problem, but keep in mind that SSD disks are expensive and they are still slower than SSD drives. For optimal performance, you want your data to be in RAM. Why not adding an extra server and use sharding.
[11:02:36] <Ceunincksken> Keksike: When you run a query with explain(), you don't get the results of your query, but you get the explanation of your query in the shell. So, you need to execute your queries directly in your shell, not in your application and see if indexes are being used and if a collection scan needs to be executed, ...
[11:02:55] <Ceunincksken> Keksike: Try to run db.<collection>.find({}).explain() in your shell and see the outcome.
[11:03:43] <Ceunincksken> Kekiske: You should use explain with "allPlansExecution" to get the most detailed stats. See: https://docs.mongodb.org/manual/reference/method/cursor.explain/
[11:05:45] <Keksike> is there any way to use this .explain inside my clojure-program and log the result of the .explain() into a file?
[11:07:29] <Ceunincksken> But, you can try to set the profiling level on your database, it might help.
[11:07:50] <Ceunincksken> run db.setProfilingLevel(2) on your database and every query you execute get's logged in a collection named 'system.profile'.
[11:07:58] <Keksike> I'm basically trying to figure out/debug why a certain operation is taking such a long time in my software
[11:08:37] <Ceunincksken> Go with 'setProfilingLevel(2)' and look at the 'system.profile' collection. It might give you a clue;
[11:09:22] <Ceunincksken> There's a setting that defines where the database write it's logging information.
[11:09:51] <newbsduser> Ceunincksken, if i have multiple clients which are sending data to single server, SSD disks (for instance 6 x 450G with raid10) can be correct solution?
[11:09:54] <Keksike> the problem with setProfilingLevel(2) is that it logs all the operations, not just the ones that I want to debug, and its a pain to filter through all of the logs just to find what im looking for. thats why I wanted to use .explain()
[11:10:33] <Ceunincksken> You can filter the system.profile on ns ans query.find to find what you are looking for.
[11:45:12] <lowbro> I have a collection with millions of records. If I configure a ttl-index in doctrine-odm on the entity for that collection, how exactly will it be applied? Will applying (and therefor removing) hundreds of thousands have a critical impact on cpu and/or memory usage?
[11:46:36] <Ceunincksken> lowbro: By default, when you build an index, it's created in the foreground, preventing all read/write operations on Mongo. You can consider creating them in the backround. See https://docs.mongodb.org/manual/tutorial/build-indexes-in-the-background/
[13:08:49] <igorwww> hello! I have a question about migrating from mongo-php-driver-legacy to mongo-php-driver, it seems like there is no equivalent to MongoCursor::timeout(), is that correct?
[13:13:59] <rom1504> Derick: spending 3min trying to find the doc for mongo-php-driver with google and failing
[13:20:38] <rom1504> igorwww: http://php.net/manual/en/mongodb-driver-manager.construct.php + https://docs.mongodb.org/manual/reference/connection-string/#connections-connection-options uri.socketTimeoutMS
[13:22:33] <igorwww> I was referring to https://github.com/mongodb/mongo-php-driver-legacy/blob/master/cursor.c#L180
[13:24:53] <rom1504> a general thing about mongodb docs : the getting started docs are nice, but when you want to find the reference, it's hard to find (if it exists at all)
[13:26:38] <igorwww> https://github.com/mongodb/mongo-php-driver#documentation https://github.com/mongodb/mongo-php-driver-legacy#documentation
[13:28:34] <rom1504> https://docs.mongodb.org/ecosystem/drivers/php/ "PHP Library API Reference" -> http://mongodb.github.io/mongo-php-library/ still a getting started
[13:38:30] <igorwww> Derick: seems like mongoc_cursor_set_max_await_time_ms() from mongo-c-driver should do the trick
[13:48:52] <mongodbuser2> Hello, I have a question that I've not been able to find the answer to. My _id is strictly increasing and no documents will ever be deleted. All queries will filter on _id with both $gt and $lt. Is there a way for me to have MongoDB use the index to find the first element, then perform a collscan until it hits the last item? As far as I understand MongoDB will use the index to retrieve all items by default, which I believe wo
[13:51:12] <mongodbuser2> I can also add that it seems that .hint($natural: 1) gives slightly better performance than when I let it use the index.
[13:51:44] <mongodbuser2> oops, sorry, here's the rest: (...) which I believe would be a bit slower (but still O(n)).
[13:52:54] <mongodbuser2> I guess what I'm asking is; Is there a way to have MongoDB exploit the fact that I know _id is strictly increasing when doing range queries?
[14:08:47] <GothAlice> hakrim: Ask your question, gatekeeper, I am not afraid! (IRC is asynchronous, so asking the question instead of asking to ask is generally more effective. ;)
[14:11:45] <hakrim> GothAlice: I'm trying to check wether my bulk has any operations in it before executing, but I'm unable to figure how to do it.
[14:15:25] <GothAlice> hakrim: For the most part the actual operations being queued up are hidden behind mangled "private" attributes.
[14:16:44] <GothAlice> http://api.mongodb.org/python/current/api/pymongo/bulk.html < not seeing that :/
[14:17:51] <GothAlice> See my screenshot; you _can_ technically pull out the queued up operations, but… it's not pretty.
[14:20:00] <GothAlice> No worries. (Being private like this means it could change without any warning between versions; I'd armour your code against the attribute not existing—try/except AttributeError.)
[14:20:18] <hakrim> It seems it could solve my problem. It also seems that I still have a lot to learn
[14:21:31] <GothAlice> hakrim: To describe how I found this: I used "ptipython" which is an improved syntax-highlighting Python REPL (pip install ipython ptpython), imported the various things, and used tab completion to explore the attributes of the object. I tracked down which attribute stored the operations by examining the code for the execute method by entering: a.execute??
[14:22:06] <hakrim> On a different topic, what is an acceptable write speed for a well made script importing from txt/json to mongo?
[14:23:11] <gablank> What clients are you using for exploring your dbs? Currently using RoboMongo but it seems it doesn't support WiredTiger.
[14:23:23] <GothAlice> JSON parsing is several orders of magnitude slower than string splitting (a la CSV), so I'd expect it to be substantially slower. On a single local node I get insertion rates in the order of two million documents per second, but that's constructing new records, not trying to translate data.
[14:24:58] <GothAlice> gablank: I use either the mongo shell (quick queries) or a Python REPL. Exploration using standard queries and aggregate queries.
[14:25:33] <hakrim> The script I'm working on average 200 docs/s. And we're supposed to be doing "Big Data" ...
[14:28:04] <gablank> When using range based sharding, does MongoDB adjust the limits automatically to balance the number of documents per shard?
[14:46:03] <hakrim> GothAlice: Thanks for the ipython thing, it's magic! No more reading obscure module source code!
[14:46:35] <GothAlice> Hehe, I recommend 'ptipython' not just 'ipython'. The former has improved tab completion display, better paste mode, etc.
[14:46:59] <GothAlice> (ptpython is the package, and if ipython is installed it can use it for the %timeit and ? helpers via running "ptipython".)
[15:44:16] <yopp> quick question about ruby driver: is it possible to get "stripped" command (without condition arguments) in instrumentation?
[15:44:50] <yopp> like {"find"=>"sites", "filter"=>{"author_id"=>{"$in"=>[BSON::ObjectId('xyz')]}}} becomes {"find"=>"sites", "filter"=>{"author_id"=>{"$in"=>[???]}}}
[16:24:10] <tantamount> If you have a collection of mixed amount of apples and bananas, e.g. ['a', 'b', 'b', 'a', 'a'] how can you aggregte how many apples and banans there are such that you end up with a document like {apple: 3, banana: 2}?
[16:24:55] <tantamount> Consider that the keys apple and banana cannot be hard-coded because there could be any number of different varieties of fruit in the collection
[16:32:04] <GothAlice> tantamount: So in this instance, the practical method is to store the counts concretely and keep the counts up-to-date at the same time you execute updates to the overall list.
[16:32:24] <mylord> correction: how do I prepend “abc” to the value of icon_url in this doc [ { items: [ { icon_url: “xyz” …
[16:33:19] <GothAlice> Well, what you're asking is only possible at the application level, unfortunately. You can't project into variably-named fields.
[16:34:02] <GothAlice> mylord: Is your intention to update those documents with the resulting value, or simply get the records returned with the text prepended (but don't update the data in the DB)?
[16:35:16] <GothAlice> Alas, there is no update operator to concatenate. During aggregate projection, however, you can https://docs.mongodb.org/manual/reference/operator/aggregation/concat/
[16:35:58] <GothAlice> You could, in theory, aggregate and $out into a new collection the updated records, then swap the collections.
[16:39:16] <GothAlice> Aggregate queries return documents transformed by the aggregation pipeline (that argument to .aggregate()). Using $out as a aggregation pipeline stage will let you save the results somewhere, but you might suffer madness if you try to $out to the same collection that it's reading from. Ref my possible solution by swapping collections.
[16:43:57] <tantamount> GothAlice: what if I only needed a unique list sorted by frequency? So the result could be ['a', 'b'] if apples were more popular than bananas or ['b', 'a'] if the reverse were true
[16:45:09] <GothAlice> tantamount: Again, by having the choices be variable, you're… somewhat limited in what you can do DB-side.
[16:47:21] <GothAlice> It would. It wouldn't produce what you were asking for, though, each type of choice would be a separate document instead of having one document with multiple fields. It'd certainly work, though.
[16:47:59] <GothAlice> So you might need to be careful to group by not just the unrolled list choice, but also the original ID in order to preserve that association, if needed.
[16:48:13] <tantamount> Right, it would be only a step in that direction, but I'm sure a series of projections and further manipulations later would be able to yield that result
[16:48:57] <GothAlice> And yeah, no, same limitation. Once exploded into distinct documents, re-combining them suffers the same "no variable field names" problem.
[16:49:57] <tantamount> Yeah, I'm sorry for the confusion, but the end result needn't be {apple: 3, banana: 2}; ['a', 'b'] would suffice, where the order is significant because it is sorted by frequency
[16:53:57] <GothAlice> mylord: And heh, sure. But you can't blame the screwdriver for being a poor hammer. (I.e. it helps to use the tool the way it was intended.) ;P
[16:55:03] <mylord> GothAlice: agreed.. #node.js and #javascript often tell me this when I ask anything mongo: http://cryto.net/~joepie91/blog/2015/07/19/why-you-should-never-ever-ever-use-mongodb/
[16:55:06] <GothAlice> MongoDB generally intends people to really, really think about their data and specifically how their data is consumed/used/queried, before actually committing data in there. Certain adjustments after-the-fact are legitimately difficult.
[16:55:48] <mylord> In that case, seems good to use the facilities of pre-configured relational DB, if you know how things should be setup from start
[16:57:21] <mylord> GothAlice: Some systems are easier to screw up. It’s good when a product is hard to screw up. Just saying
[16:57:43] <GothAlice> The counter-points to those initial "it's evil" points are trivial. 1. RTFM and learn how to use writeConcern. 2. RTFM. 3. Plenty of counter examples of insane performance. (My own being ~2 million record/second bidirectionally confirmed RPC calls.) 3. Forces? It does no such thing, that's just a developer experience thing. 4. The referenced article for "locking issues" describes locking issues in MySQL, not MongoDB.
[16:58:01] <mylord> I don’t like node either. Wish I used C# or Scala.. I don’t need asynchronous EVERYWHERE.. hard to code it
[16:58:30] <GothAlice> mylord: I think a key point in an article which actually bothers to cite references is that the references should say the things the summary suggests they do. :|
[17:00:44] <GothAlice> Also "nightmare to scale and maintain". XD My servers average three years uptime, or did until recently darn you Rackspace maintenance, and I've been storing greater than 30 terabytes of data in MongoDB for many years. (Since 1.2 or 1.4, can't really remember at this point, though back then it was more like 20TB.)
[17:01:29] <GothAlice> mylord: Lesson of the day: check angry blog post references, and take them with a grain of salt in that the entire thing may be because they didn't read. ;P
[17:03:22] <GothAlice> The articles points about using the right tool for the job are spot-on, though. If you're using Mongoose, use something else. If you're storing highly relational data? Use a relational database. (Similarly, for the love of the gods, if you're storing a graph use a graph database… faking graphs is even harsher than faking joins!)
[17:03:33] <mylord> ya, agreed.. Just example of easy system: I didn’t know C#.. I jumped on C# / SQL project.. everything took just minutes to write correctly the first time.. Node +Mongo.. tool me so long to get things workign as intended, and continues that way.. I probably need to read more, but there’s a lot to read, and all kinds of special tricks to get things done.. SQL big queries can be confusing, but it all follows a small set of keywords in
[17:04:21] <mylord> Ya, right tool for right task for sure.. I think there are some cases when Document storage makes sense.. tho most problems are relational in nature
[17:04:52] <GothAlice> MongoDB is a very different way of thinking about your data. If you have been trained to think of data like an Excel spreadsheet, you've basically already lost. ;)
[17:05:43] <mylord> I do get things done with mongo, and understand it’s different.. It’s very cool in a way
[17:07:17] <GothAlice> mylord: The most recent example of awesome I use is the document validation system in MongoDB. People often complain Mongo doesn't have schemas. Well, they're not exactly correct. https://docs.mongodb.org/manual/core/document-validation/
[17:07:56] <GothAlice> Where a relational/SQL database has a fixed set of columns for a table, MongoDB instead lets you configure an _insanely expressive_ document query which must match for a document to be "valid". This lets you set up conditionals, for example.
[17:08:11] <GothAlice> (I.e. "if this user is an administrator, their e-mail address must end in @companyname.com".)
[17:09:57] <freeone3000> So I have a monitoring agent which is complaining that: https://gist.github.com/freeone3000/0c00776aed960b617d39 . However, I for sure have hosts in my MMS group in the config, and the host is being counted as "down" from MMS (even though application servers can hit it fine), so there's no other monitoring agent.
[17:10:50] <igorwww> hello again :) I am running into an error with the mongoc driver for php when trying to set a $comment https://gist.github.com/igorwwwwwwwwwwwwwwwwwwww/ce8a9695e594f0d38004
[17:11:01] <mylord> that’s cool.. I guess if you’re *really* good with it, it provides more power.. but common things like queries across documents, complex updates, etc, that are easy in SQL, take some time to understand how to do right in Mongo.. hard to figure out the correct data model, and have to think about it.. in SQL there’s sort of 1 correct way, orthogonal
[17:11:17] <igorwww> I'm not quite sure where to put $query or $comment in this case to make it work
[17:11:44] <mylord> compound that with asynch in node, and it’s been sort of cumbersome to me in some ways, but nice in others
[17:13:40] <freeone3000> Ah. It looks like the monitoring agent for my first cluster is also attempting to monitor my second cluster. That's not going to fly. What's the proper way to segregate MMS monitoring agents for multiple replication sets? I want one per replset because they have different auth
[17:13:53] <igorwww> I found test_comment() in tests/test-mongoc-collection-find.c; but I was not able to deduce from this where the $query should go with a findandmodify
[17:17:30] <tantamount> GothAlice: I still don't get it :( If I sort by fruit, it will just sort lexically. How do I know how many there are of each?
[17:17:34] <GothAlice> mylord: Queries across documents… are relational thinking. You instead structure your documents to meet query requirements: http://www.javaworld.com/article/2088406/enterprise-java/how-to-screw-up-your-mongodb-schema-design.html
[17:18:23] <mylord> i do some of that tho.. nesting them as i need them.. it took a lot of thinking, a lot of changing, at first, esp on first 2 mongodb projects
[17:19:43] <mylord> with mongodb, i setup things as i need them, mostly for presenting the client, already nested with all the data i need.. but then sometimes i’d be like.. oh, i should have structured differently
[17:20:01] <mylord> but .. anyhow with sql, you need to think how the API will structure the JSON.. so same problem
[17:20:15] <mylord> with mongo it’s nice you’re already structuring things how you want to e.g. present them
[17:20:57] <GothAlice> tantamount: You need to unwind, then sum, then re-group. I.e. [ {$unwind: "$fruit"}, {$group: {_id: {oid: "$_id", fruit: "$fruit"}, {count: {$sum: 1}}}}, {$sort: {"_id.oid": 1, "count": 1}}, {$group: {_id: "$_id.oid", "orderedfruit": {$push: "$_id.fruit"}}} ]
[17:21:37] <mylord> no translation needed to/from sql, tho maybe doc engine is a bit harder to work with, generally speaking, for complex things on the data itself, in the interim operations
[17:22:37] <GothAlice> "Throw data at the wall to see what sticks" doesn't work any better in SQL than it does in MongoDB, mylord. ¬_¬ If you don't plan, you plan for failure.
[17:23:47] <GothAlice> (A six-way join with sub-queries is as much a bad sign in SQL as almost any type of relational-ness is a bad sign in MongoDB. ;)
[17:24:15] <mylord> It’s easier for me to plan in SQL.. the solutions seem totally obvious to me. So it’s a nice challenge now.
[17:24:55] <GothAlice> I can highly recommend reading through that link I gave you last; it might help give a summary of the major differences and their approaches.
[17:25:28] <mylord> Have 10 bugs to fix tonight.. will try to not watch politics instead of read tech articles
[17:26:10] <GothAlice> Left outer, specifically: https://docs.mongodb.org/manual/reference/operator/aggregation/lookup/
[17:26:26] <mylord> Cool. I just feel so comfy in SQL.. anything you need, I can do, even if I didn’t know how before. Very simple. I’m starting to get there with MongoDB as well.
[17:27:13] <mylord> Sucky with SQL is all that DB setup time. In MS SQL they have nice tools, but PostgreSQL, not so much
[17:52:41] <freeone3000> How can I tell my MMS Monitoring Agent which hosts I want it to monitor? MMS is trying to use one monitoring agent for all clusters - I want one agent for one cluster, and another agent for a second cluster.
[17:55:11] <ParisHolley> production cluster crashed. can anyone provide any idea why PID’s are going into the 100K+ mark? i boosted my max pid to help mitigate the issue, but it just keeps draining more cpu and the server is unresponsive. granted, I have a ton of connections coming in (~30K), but this has never been an issue in the past
[17:56:08] <saira_123> hi Professionals plotting mongostat output , can someone tell me if it is possible to plot everything in one graph?
[17:58:24] <saira_123> hi Professionals plotting mongostat output , can someone tell me if it is possible to plot everything in one graph?
[17:58:43] <saira_123> e.g insert query update delete getmore command % dirty % used flushes vsize res qr|qw ar|aw netIn netOut conn time *0 4867 *0 *0 0 1|0 0.2 20.4 0 3.2G 2.6G 0|0 2|0 940k 286m 4 2016-02-11T22:52:05+05:00 *0 5905 *0 *0 0 1|0 0.2 20.4 0 3.2G 2.6G 0|1 2|0 1m 344m 4 2016-02-11T22:52:06+05:00
[19:47:03] <uuanton> hi, when secondaries at startup2 state and traffic come to them what happened ?
[19:53:54] <GothAlice> uuanton: I'm… not entirely sure what lead you to ask that question. Until it's actually in a running state (not startup or init-sync) it won't answer queries at all.
[19:54:28] <GothAlice> So it must go to a different secondary (one already started and waiting) or the primary.
[19:55:33] <uuanton> oh glad you here Alice. There was a fatal on few secondaries i clear data folder and resync all secondaries to primary
[19:56:24] <uuanton> only primary up to date right now and application doesn't work because website suppose to &readPreference=secondaryPreferred
[19:56:46] <GothAlice> Well, secondaryPreferred should fall back on the primary; only secondaryOnly should actually fail if there are no secondaries. :/
[23:08:39] <GothAlice> CustosLimen: Generally a good idea to look at the server logs to see what's going on, too.
[23:13:49] <GothAlice> You regularly run parallel, aggressively batched, unlimited rate multi-collection bulk inserts in production?
[23:14:55] <CustosLimen> GothAlice, look, I'm not saying from this I can assume it will happen in production use cases - but without understaning why this is happening I will remain worried
[23:26:43] <beryl> now that mongodb is using wired tiger and that there is no limit on the number of collections, is it a bad pattern to create something like one collection per user?
[23:27:15] <GothAlice> beryl: Consider: do you ever want to be able to ask a question across multiple user's data at once?
[23:28:05] <GothAlice> Any time I hear someone wanting to do something just because some limit has been lifted I'm reminded of: http://s.webcore.io/d41m
[23:29:09] <GothAlice> And I mean question in a more general sense than "supply answers to your users questions". I mean the company's questions, too.
[23:31:47] <GothAlice> So… how does adding a field to the data referencing the user who owns it not satisfy that requirement?
[23:33:42] <GothAlice> (Dynamic fields being a _terrible_ choice with _substantial_ ramifications, but that's a different issue than tracking who owns what.)
[23:33:45] <beryl> it totally statisfies it, but out of curiosity and open mindedness, is it reasonable to create a collection by user and if not why exactly? that's the question that i can't an answer
[23:34:48] <GothAlice> Split index building, so extra overhead there. On-disk stripe allocation… so, more overhead there. It gives a false sense of data isolation; you can, without some careful rain dancing, easily switch from one collection to another, making the isolation entirely virtual.
[23:35:38] <GothAlice> And you lose on any ability to correlate data. Even if your users don't need it, your company might want to be able to ask questions of the data that goes beyond "one user at a time". I.e. if you need to find out which user has X data, instead of one query that'll be indexed and _blindingly_ fast, you'll need to issue N queries for N of your user count.
[23:36:18] <GothAlice> For real separation, and the additional hardware costs that actually involves, you can use sharding with a sharding key per user.
[23:36:32] <GothAlice> Er, rather, a sharding key referencing a unique value per user. Like their ID, or username.
[23:37:47] <beryl> I was afraid that because one user will have a lot of data, it would slow down every other users
[23:38:13] <GothAlice> With proper indexes, i.e. including the owner field as a first field in any compound index, it'll stay fast.
[23:40:54] <beryl> ok, so sharding by user id, multi-key index on the user id and the _id of the record?
[23:42:52] <GothAlice> Sharding by owner is one form of "true" data isolation (that still allows for broader queries), including the owner as a prefix to your compound indexes keeps the queries fast (since basically every query will need to include the owner). "True" isolation that doesn't allow for broader queries is to split into multiple databases and use a connection pool per database, with a MongoDB-level user account per database.
[23:43:41] <GothAlice> If you're doing a lookup by _id, then the owner check is a security measure, not a method of making the query more specific. _id is automatically indexed.
[23:44:32] <GothAlice> For example, if you have a "name" field (just as an example ;) which you want to index, you'd actually index {owner: 1, name: 1} — then queries like db.foo.find({owner: "GothAlice", name: "Bob Dole"}) will be fast.
[23:44:42] <beryl> I don't need true isolation, i just don't want a user with 100 000 records to slow down users with 10
[23:45:26] <GothAlice> beryl: We generate a million log events per month, each user only seeing a fraction of those on their dashboard. Such concern is unfounded, generally.
[23:46:03] <GothAlice> (As long as you have good indexes. Performance, regardless of this isolation problem, almost always comes down to the indexes.)
[23:48:05] <GothAlice> beryl: Alice's Law 146: Optimization without measurement is by definition premature. [https://gist.github.com/amcgregor/9f5c0e7b30b7fc042d81] Turning a collection structured "naturally" (with owner references) into a sharded collection or other arrangement is relatively trivial. Merging a metric ton of individual collections into one after realizing it just makes everything harder… is much, much more painful, too.
[23:52:06] <GothAlice> I totally need a bot so I can just reference a law by number and have it pull it up. XD