I have a flow like this:
I have a Worker that's processing a "large" batch (say, 1M records) and storing the results in Mongo.
Once the batch is complete, a notification message is sent to Publish, which then pulls all the records from Mongo for final publication.
Let's say the Worker write process is done, i.e. it has sent all 1M records to Mongo through a driver. Mongo is "eventually consistent" so I'm not 100% guaranteed all records are written to physical storage at the time the Notify Publish happens.
When Publish does a 'find' and gets a cursor on the collection holding the batch records, is the cursor smart enough to handle the eventual consistency?
So in practical terms let's imagine 750,000 records are actually physically written by Mongo when Notify Publish happens and Publish does its find. Will the cursor traverse 750,000 records and stop or will it block or otherwise handle the remaining 250,000 as they're eventually written to disk (which presumably is very likely to happen while publishing of the first 750K)?
As #BlakesSeven already noted in the comments, "eventual consistency" refers to the fact that in a replicated environment, when a write is finished on the primary, it will only be written to the secondaries eventually. You can modify this behavior at the cost of reduced write performance by setting the write concern to > 1. Setting it to "majority" basically guarantees that a write operation is durable even in case of a failover – though at a (in some cases) drastically reduced performance.
In general here is what happens when you do a write (simplified) with journaling enabled:
The operation is checked for being syntactically correct.
The query optimizer kicks in and does his stuff. (Irrelevant for this question, so I spare the details).
The write operation is applied to the in memory representation of the data set called "private view".
Every commitIntervalMs, the private view is synced to the journal, with a median of 15 or 50ms, depending on the write concern.
On sync, the operation is applied to the shared view. Iirc, this is the point where a new connection would be provided with the new data.
So in order to ensure that the data will be readable by the new connection, simply delay the publish notification by commitIntervalMs + 1, which, given your batch size, is hardly noticeable.
I've been read a lot about MongoDB recently, but one topic I can't find any clear material on, is how data is written to the journal and oplog.
So this is what I understand of the process so far, please correct me where I'm wrong
A client connect to mongod and performs a write. The write is stored in the socket buffer
When Mongo is available (not sure what available means at this point), data is written to the journal?
The mongoDB docs then say that writes every 60 seconds are flushed from the journal onto disk. By this I can only assume this mean written to the primary and the oplog. If this is the case, how to writes appear earlier than the 60 seconds sync interval?
Some time later, secondaries suck data from the primary or their sync source and update their oplog and databases. It seems very vague about when exactly this happens and what delays it.
I'm also wondering if journaling was disabled (I understand that's a really bad idea), at what point does the oplog and database get updated?
Lastly I'm a bit stumpted at which points in this process, the write locks get created. Is this just when the database and oplog are updated or at other times too?
Thanks to anyone who can shed some light on this or point me to some reading material.
Simon
Here is what happens as far as I understand it. I simplified a bit, but it should make clear how it works.
A client connects to mongo. No writes done so far, and no connection torn down, because it really depends on the write concern what happens now.Let's assume that we go with the (by the time of this writing) default "acknowledged".
The client sends it's write operation. Here is where I am really not sure. Either after this step or the next one the acknowledgement is sent to the driver.
The write operation is run through the query optimizer. It is here where the acknowledgment is sent because with in an acknowledged write concern, you may be returned a duplicate key error. It is possible that this was checked in the last step. If I should bet, I'd say it is after this one.
The output of the query optimizer is then applied to the data in memory Actually to the data of the memory mapped datafiles, to the memory mapped oplog and to the journal's memory mapped files. Queries are answered from this memory mapped parts or the according data is mapped to memory for answering the query. The oplog is read from memory if present, too.
Every 100ms in general the journal is synced to disk. The precise value is determined by a number of factors, one of them being the journalCommitInterval configuration parameter. If you have a write concern of journaled, the driver will be notified now.
Every syncDelay seconds, the current state of the memory mapped files is synced to disk I think the journal is truncated to the entries which weren't applied to the data yet, but I am not too sure of that since that it should basically never happen that data in the journal isn't yet applied to the current data.
If you have read carefully, you noticed that the data is ready for the oplog as early as it has been run through the query optimizer and was applied to the files mapped into memory. When the oplog entry is pulled by one of the secondaries, it is immediately applied to it's data of the memory mapped files and synced in the disk the same way as on the primary.
Some things to note: As soon as the relatively small data is written to the journal, it is quite safe. If a node goes down between two syncs to the datafiles, both the datafiles and the oplog can be restored from their last state in the datafiles and the journal. In general, the maximum data loss you can have is the operations recorded into the log after the last commit, 50ms in median.
As for the locks. If you have written carefully, there aren't locks imposed on a database level when the data is synced to disk. Write locks may be created in order to assure that only one thread at any given point in time modifies a given document. There are other write locks possible , but in general, they should be rather rare.
Write locks on the filesystem layer are created once, though only implicitly, iirc. During application startup, a lock file is created in the root directory of the dbpath. Any other mongod instance will refuse to do any operation on those datafiles while a valid lock exists. And you shouldn't either ;)
Hope this helps.
Even if journaling is on, is there still a chance to lose writes in MongoDB?
"By default, the greatest extent of lost writes, i.e., those not made to the journal, are those made in the last 100 milliseconds."
This is from Manage Journaling, which indicates you could lose writes made since the last time the journal was flushed to disk.
If I want more durability, "To force mongod to commit to the journal more frequently, you can specify j:true. When a write operation with j:true is pending, mongod will reduce journalCommitInterval to a third of the set value."
Even in this case, it looks like flushing the journal to disk is asynchronous so there is still a chance to lose writes. Am I missing something about how to guarantee that writes are not lost?
Posting a new answer to clean this up. I performed tests and read the source code again and I'm sure the irritation comes from an unfortunate sentence in the write concern documentation. With journaling enabled and j:true write concern, the write is durable, and there is no mysterious window for data loss.
Even if journaling is on, is there still a chance to lose writes in MongoDB?
Yes, because the durability also depends on the individual operations write concern.
"By default, the greatest extent of lost writes, i.e., those not made to the journal, are those made in the last 100 milliseconds."
This is from Manage Journaling, which indicates you could lose writes made since the last time the journal was flushed to disk.
That is correct. The journal is flushed by a separate thread asynchronously, so you can lose everything since the last flush.
If I want more durability, "To force mongod to commit to the journal more frequently, you can specify j:true. When a write operation with j:true is pending, mongod will reduce journalCommitInterval to a third of the set value."
This irritated me, too. Here's what it means:
When you send a write operation with j:true, it doesn't trigger the disk flush immediately, and not on the network thread. That makes sense, because there could be dozens of applications talking to the same mongod instance. If every application were to use journaling a lot, the db would be very slow because it's fsyncing all the time.
Instead, what happens is that the 'durability thread' will take all pending journal commits and flush them to disk. The thread is implemented like this (comments mine):
sleepmillis(oneThird); //dur.cpp, line 801
for( unsigned i = 1; i <= 2; i++ ) {
// break, if any j:true write is pending
if( commitJob._notify.nWaiting() )
break;
// or the number of bytes is greater than some threshold
if( commitJob.bytes() > UncommittedBytesLimit / 2 )
break;
// otherwise, sleep another third
sleepmillis(oneThird);
}
// fsync all pending writes
durThreadGroupCommit();
So a pending j:true operation will cause the journal commit thread to commit earlier than it normally would, and it will commit all pending writes to the journal, including those that don't have j:true set.
Even in this case, it looks like flushing the journal to disk is asynchronous so there is still a chance to lose writes. Am I missing something about how to guarantee that writes are not lost?
The write (or the getLastError command) with a j:true journaled write concern will wait for the durability thread to finish syncing, so there's no risk of data loss (as far as the OS and hardware guarantee that).
The sentence "However, there is a window between journal commits when the write operation is not fully durable" probably refers to a mongod running with journaling enabled that accepts a write that does NOT use the j:true write concern. In that case, there's a chance of the write getting lost since the last journal commit.
I filed a docs bug report for this.
Maybe. Yes, it waits for the data to be written, but according to the docs there's a 'there is a window between journal commits when the write operation is not fully durable', whatever that is. I couldn't find out what they refer to.
I'm leaving the edited answer here, but I reversed myself back-and-forth, so it's a bit irritating:
This is a bit tricky, because there are a lot of levers you can pull:
Your MongoDB setup
Assuming that journaling is activated (default for 64 bit), the journal will be committed in regular intervals. The default value for the journalCommitInterval is 100ms if the journal and the data files are on the same block device, or 30ms if they aren't (so it's preferable to have the journal on a separate disk).
You can also change the journalCommitInterval to as little as 2ms, but it will increase the number of write operations and reduce overall write performance.
The Write Concern
You need to specify a write concern that tells the driver and the database to wait until the data is written to disk. However, this won't wait until the data has been actually written to the disk, because that would take 100ms in a bad-case scenario with the default setup.
So, at the very best, there's a 2ms window where data can get lost. That's insufficient for a number of applications, however.
The fsync command forces a disk flush of all data files, but that's unnecessary if you use journaling, and it's inefficient.
Real-Life Durability
Even if you were to journal every write, what is it good for if the datacenter administrator has a bad day and uses a chainsaw on your hardware, or the hardware simply disintegrates itself?
Redundant storage, not on a block device level like RAID, but on a much higher level is a better option for many scenarios: Have the data in different locations or at least on different machines using a replica set and use the w:majority write concern with journaling enabled (journaling will only apply on the primary, though). Use RAID on the individual machines to increase your luck.
This offers the best tradeoff of performance, durability and consistency. Also, it allows you to adjust the write concern for every write and has good availability. If the data is queued for the next fsync on three different machines, it might still be 30ms to the next journal commit on any of the machines (worst case), but the chance of three machines going down within the 30ms interval is probably a millionfold lower than the chainsaw-massacre-admin scenario.
Evidence
TL;DR: I think my answer above is correct.
The documentation can be a little irritating, especially with regards to wtimeout, so I checked the source. I'm not an expert on the mongo source, so take this with a grain of salt:
In write_concern.cpp, we find (edited for brevity):
if ( cmdObj["j"].trueValue() ) {
if( !getDur().awaitCommit() ) {
// --journal is off
result->append("jnote", "journaling not enabled on this server");
} // ...
}
else if ( cmdObj["fsync"].trueValue() ) {
if( !getDur().awaitCommit() ) {
// if get here, not running with --journal
log() << "fsync from getlasterror" << endl;
result->append( "fsyncFiles" , MemoryMappedFile::flushAll( true ) );
}
Note the call MemoryMappedFile::flushAll( true ) if fsync is set. This call is clearly not in the first branch. Otherwise, durability is handled on a sepate thread (relevant files prefixed dur_).
That explains what wtimeout is for: it refers to the time waiting for slaves, and has nothing to do with I/O or fsync on the server.
Journaling is for keeping the data on a particular mongod in a consistent state, even in case of chainsaw madness, however with client settings through writeconcern it can be used to force out durability. About write concern DOCS.
There is an option, j:1, which you can read about here which ensures that the particular write operation waits for acknowledge till it is written to the journal file on disk (so not just in the memory map). However this docs says the opposite. :) I would vote for the first case it makes me feel more comfortable.
If you run lots of commands with such option mongodb will adapt the size of the commit interval of the journal to speed things up, you can read about it here: DOCS this one you also mentioned and as others already said that you can specify an interval between 2-300ms.
Durability is much more ensured in my opinion over the w:2 option while if the update/write operation is acknowledged by two members in a replicaset it is really unlikely to lose both in the same minute (datafile flush interval), but not impossible.
Using both options will cause the situation that when the operation is acknowledged by the database cluster it will reside in memory at two different boxes and on one it will be in a consistent recoverable disk place too.
Generally lost writes are an issue in every system where there is buffering/caching/delayed-write involved between a system's runtime and a permanent (non-volatile) storage, even at the OS level (for example write-behind caching). So there is always a chance to lose writes, even if your concrete provider (MongoDB) provides functionality for transaction durability it's the underlying OS that is responsible for ultimately writing the data, and even then there is caching at the device level... And that's just the lower levels, making the system highly concurrent, distributed and performant only makes matters worse.
In short there is no absolute durability, only practical/eventual/hope-for-the-best durability especially with a NoSQL storage like Mongo, which isn't primarily made for consistency and durability in the first place.
I would have to agree with Sammaye that journoualing has little to do with durability. However, if you want to get an answer to whether you can really trust mongodb to store your data with good consistency, then I would suggest that you read this blog post. There is a reply from 10gen regarding that post, and a reply from the author to the 10gen post. I would suggest that you read into it to make an educated decision. It took me some time to understand all the details on my own, but this post has the basics covered.
The response to the blog post was given here by 10gen, the company that makes mongodb.
And the response to the response was given by the professor on this post.
It explains a lot about how Mongodb can shard data, how it actually functions, and the performance hits it takes if you add on extra safety locks. I strongly want to say that these three writings are the best thing out there, and by far the most comprehensive things out there that talk about the benefits and drawbacks of mongodb, if you think its one sided, look at the comments, and also see what people had to say, because if something received a reply from the company that made the software, then it must have made some good points atleast.
I have been reading a few articles on MongoDB fault tolerance, and some people complain it can never really be achieved with MongoDB (like in this article: http://hackingdistributed.com/2013/01/29/mongo-ft/), and this got be confused.
Can someone confirm (and if possible show me the appropriate docs) that using the Write Concern "Journal + Majority" is enough to make sure that 100% of the writes that were reported as success by my driver are durably written and won't be lost even if any replica fails just after the write?
I'm talking about a 3 replica setup. I'm ok with the system no longer accepting writes in case of failure, but when a write is reported as successful by the driver, I need it to be durably committed (regardless of the number of replica failing after that).
Right so if you choose a journal write you are basically ensuring the write has made it to disk of a single node. If you choose to do a majority write, you are ensuring that the write has made it to memory of at least x number of nodes in your replica set.
By default, mongodb will flush from memory to journal every 100ms. By having your replica nodes on different machines (physical or virtual), ideally in different data centres, you are very unlikely to ever see ALL nodes in a geographically ditributed replica set go down within the same 100ms before one gets to disk.
Alternatively to guarantee that write made it to disk of a single node - use journal write.
I have safe mode turned off in my MongoDB database because none of the data being written is absolutely 100% mission critical and the gain in insertion speed is very important, but I would really prefer if all of the data is written to the database.
My understanding is that with journaling turned on and safe mode turned off, if the server crashes in the 100ms between when a write request is received and the data is output to the journal, the data can be lost.
If the data is successfully written to the journal, is it a pretty safe bet, even if the database is lagging due to heavy load, that the data will end up in the database when the database catches up and is able to process what's in the journal? Or is my understanding of what the journal does flawed? Are there any other circumstances under which inserted data may be lost?
What happens if I update a document a fraction of a second before another process attempts to read it, but the changes haven't been committed to the collection yet? Will the read block until the insert has completed?
The read will only be blocked until the insert has completed if the read is requested on the same connection as the write. There is no guarantee that once the data is written, it will be immediately visible to other connections unless proper getLastError is used.
Data is processed and put in memory mapped region of data before journaling. However, it may not be "fsynced" to disk as often as journaling. This means that even though the load is high, the data should be eventually updated and become visible to other connections. Journaling data is only used to restore durability when mongod instance unexpectedly crashes.
Your data may be lost due to network interruption, disk corruption, index dup, etc.