Been working with MongoDB for a while and today I had a doubt while discussing with a colleague.
The thing is that when you create an index in MongoDB, the collection is processed and the index is built.
The index is updated within insertion and deletion of documents so I don't really see the need to run a rebuild index operation (which drops the index and then rebuild it).
According to MongoDB documentation:
Normally, MongoDB compacts indexes during routine updates. For most
users, the reIndex command is unnecessary. However, it may be worth
running if the collection size has changed significantly or if the
indexes are consuming a disproportionate amount of disk space.
Does someone has had the need of running a rebuild index operation that worth it?
As per the MongoDB documentation, there is generally no need to routinely rebuild indexes.
NOTE: Any advice on storage becomes more interesting with MongoDB 3.0+, which introduced a pluggable storage engine API. My comments below are specifically in reference to the default MMAP storage engine in MongoDB 3.0 and earlier. WiredTiger and other storage engines have different storage implementations for data & indexes.
There may be some benefit in rebuilding an index with the MMAP storage engine if:
An index is consuming a larger than expected amount of space compared to the data. Note: you need to monitor historical data & index size to have a baseline for comparison.
You want to migrate from an older index format to a newer one. If a reindex is advisible this will be mentioned in the upgrade notes. For example, MongoDB 2.0 introduced significant index performance improvements so the release notes include a suggested reindex to the v2.0 format after upgrading. Similarly, MongoDB 2.6 introduced 2dsphere (v2.0) indexes which have a different default behaviour (sparse by default). Existing indexes are not rebuilt after index version upgrades; the choice of if/when to upgrade is left to the database administrator.
You have changed the _id format for a collection to or from a monotonically increasing key (eg. ObjectID) to a random value. This is a bit esoteric, but there's an index optimisation that splits b-tree buckets 90/10 (instead of 50/50) if you are inserting _ids that are always increasing (ref: SERVER-983). If the nature of your _ids changes significantly, it may be possible to build a more efficient b-tree with a re-index.
For more information on general B-tree behaviour, see: Wikipedia: B-tree
Visualising index usage
If you're really curious to dig into the index internals a bit more, there are some experimental commands/tools you can try. I expect these are limited to MongoDB 2.4 & 2.6 only:
indexStats command
storage-viz tool
While I don't know the exact technical reasons why, in MongoDB, I can make some assumptions about this, based on what I know about indexing from other systems and based on the documentation that you quoted.
The General Idea Of An Index
When moving from one document to the next, in the full document collection, there is a lot of wasted time and effort skipping past all the data that doesn't need to be dealt with. If you're looking for document with id "1234", having to move through 100K+ of each document makes it slow
Rather than having to search through all of the content of each document in the collection (physically moving the disk read heads, etc), an index makes this fast. It's basically a key/value pair that gives you the id and the location of that document. MongoDB can quickly scan through all of the id's in the index, find the locations of the documents that it needs, and go load them directly.
Allocating File Size For An Index
Indexes take up disk space because they are basically a key/value pair stored in a much smaller location. If you have a very large collection (large number of items in the collection) then your index grows in size.
Most operating systems allocate chunks of disk space in certain block sizes. Most database also allocate disk space in large chunks, as needed.
Instead of growing 100K of file size when 100K of documents are added, MongoDB will probably grow 1MB or maybe 10MB or something - I don't know what the actual growth size is. In SQL Server, you can tell it how fast to grow, and MongoDB probably has something like that.
Growing in chunks give the ability to 'grow' the documents in to the space faster because the database doesn't need to constantly expand. If the database now has 10MB of space already allocated, it can just use that space up. It doesn't have to keep expanding the file for each document. It just has to write the data to the file.
This is probably true of collections and indexes for collections - anything that is stored on disk.
File Size And Index Re-Building
When a large collection has a lot of documents added and removed, the index becomes fragmented. index keys may not be in order because there was room in the middle of the index file and not at the end, when the index needed to be built. Index keys may have a lot of space in between them, as well.
If there are 10,000 items in the index, and # 10,001 needs to be inserted, it may be inserted in the middle of the index file. Now the index needs to re-build itself to put everything back in order. This involves moving a lot of data around, to make room at the end of the file and put item # 10,001 at the end.
If the index is constantly being thrashed - lots of stuff removed and added - it's probably faster to just grow the index file size and always put stuff at the end. this is fast to create the index, but leaves empty holes in the file where old things were deleted.
If the index file has empty space where deleted things used to be, this is wasted effort when reading the index. The index file has more movement than needed, to get to the next item in the index. So, the index repairs itself... which can be time consuming for very large collections or very large changes to a collection.
Rebuild For A Large Index File
It can take a lot of disk access and I/O operations to correctly compact the index file back down to a reasonable size, with everything in order. Move out of place items to temp location, free up space in right spot, move them back. Oh by the way, to free up space, you had to move other items to temp location. It's recursive and heavy-handed.
Therefore, if you have a very large number of items in a collection and that collection has items added and removed on a regular basis, the index may need to be rebuilt from scratch. Doing this would wipe the current index file and rebuild from the ground up - which is probably going to be faster than trying to do thousands of moves inside of the existing file. Rather than moving things around, it just writes them sequentially, from scratch.
Large Change In Collection Size
Giving everything I'm assuming above, a large change in the collection size would cause this kind of thrashing. If you have 10,000 documents in the collection and you delete 8,000 of them... well, now you have empty space in your index file where the 8,000 items used to be. MongoDB needs to move the remaining 2,000 items around in the physical file, to rebuild it in a compact form.
Instead of waiting around for 8,000 empty spaces to be cleaned up, it might be faster to rebuild from the ground up with the remaining 2,000 items.
Conclusion? Maybe?
So, the documentation that you quoted is probably going to deal with "big data" needs or high thrashing collections and indexes.
Also keep in mind that I'm making an educated guess based on what I know about indexing, disk allocation, file fragmentation, etc.
My guess is that "most users" in the documentation, means 99.9% or more of mongodb collections don't need to worry about this.
MongoDB specific case
According to MongoDB documentation:
The remove() method does not remove the indexes
So if you delete documents from a collection you are wasting disk space unless you rebuild the index for that collection.
Related
From this book: https://www.oreilly.com/library/view/50-tips-and/9781449306779/ch01.html
Specifically about "Tip #7: Pre-populate anything you can".
It claims that pre-populating data is better because
MongoDB does not need to find space for them. It merely updates the values you’ve already entered, which is much faster
Is there any truth in this? I've checked the MongoDB manual about data modeling and it does not mention anything about this.
The other tips also does not cite any sources so I'm wondering if there are any basis in these tips
Is there any truth in this?
Yes, if you're using MMAPv1 storage engine.
https://docs.mongodb.com/manual/core/write-performance/#document-growth-and-the-mmapv1-storage-engine
Some update operations can increase the size of the document; for instance, if an update adds a new field to the document.
For the MMAPv1 storage engine, if an update operation causes a document to exceed the currently allocated record size, MongoDB relocates the document on disk with enough contiguous space to hold the document. Updates that require relocations take longer than updates that do not, particularly if the collection has indexes. If a collection has indexes, MongoDB must update all index entries. Thus, for a collection with many indexes, the move will impact the write throughput.
I've learned a lot of things about indexing and finding some stuff from
here.
Indexes support the efficient execution of queries in MongoDB. Without indexes, MongoDB must perform a collection scan, i.e. scan every document in a collection, to select those documents that match the
query statement. If an appropriate index exists for a query, MongoDB can use the index to limit the number of documents it must inspect.
But i still have some questions:
While Creating index using (createIndex), is the Record always stored in
RAM?
Is every time need to create Index Whenever My application
is going to restart ?
What will Happen in the case of default id (_id). Is always Store in RAM.
_id Is Default Index, That means All Records is always Store in RAM for particular collections?
Please help me If I am wrong.
Thanks.
I think, you are having an idea that indexes are stored in RAM. What if I say they are not.
First of all we need to understand what are indexes, indexes are basically a pointer to tell where on disk that document is. Just like we have indexing in book, for faster access we can see what topic is on which page number.
So when indexes are created, they also are stored in the disk, But when an application is running, based on the frequent use and even faster access they get loaded into RAM but there is a difference between loaded and created.
Also loading an index is not same as loading a collection or records into RAM. If we have index loaded we know what all documents to pick up from disk, unlike loading all document and verifying each one of them. So indexes avoid collection scan.
Creation of indexes is one time process, but each write on the document can potentially alter the indexing, so some part might need to be recalculating because records might get shuffled based on the change in data. that's why indexing makes write slow and read fast.
Again think of as a book, if you add a new topic of say 2 pages in between the book, all the indexes after that topic number needs to be recalculated. accordingly.
While Creating index Using (createIndex),Is Record always store in RAM
?.
No, records are not stored in RAM, while creating it sort of processes all the document in the collection and create an index sheet, this would be time consuming understandably if there are too many documents, that's why there is an option to create index in background.
Is every time need to create Index Whenever My application is going to
restart ?
Index are created one time, you can delete it and create again, but it won't recreated on the application or DB restart. that would be insane for huge collection in sharded environment.
What will Happen in the case of default id (_id). Is always Store in
RAM.
Again that's not true. _id comes as indexed field, so index is already created for empty collection, as when you do a write , it would recalculate the index. Since it's a unique index, the processing would be faster.
_id Is Default Index, That means All Records is always Store in RAM for particular collections ?
all records would only be stored in RAM when you are using in-memory engine of MongoDB, which I think comes as enterprise edition. Due to indexing it would not automatically load the record into RAM.
To answer the question from the title:
MongoDB indexes use a B-tree data structure.
source: https://docs.mongodb.com/manual/indexes/index.html#b-tree
Say we would like to store keywords as an array in mongodb and index them for faster look up, how does common performance issue with large array indexes apply?
{
text: "some text",
keyword: [ "some", "text" ]
}
Depending on the length of text, the keyword set might get quite large. If we set index as background, does it mitigate the slow down during document insertion? We are unlikely going to modify a keyword once it's created.
PS: we know about the experimental text search in mongodb, but some of our texts are not in the list of supported languages (think CJK), so we are considering a simple home-brew solution.
The issue that is mentioned in the "common performance issue" link that you point talks about modifying the array later. If you keep pushing elements to an array, MongoDB will need to move the document on disk. When it moves a document on disk, all the indexes that point to the moved document also need to be updated.
In your case, you will not be modifying the arrays, so there is no performance degradation due to moving documents around.
I don't think you even need to turn on background indexes. This is a feature that is meant for relieving locking on the database when you add an index to an already existing collection. Depending on the collection, the index build can take a long time and hence you would benefit from sacrificing some index building time for not-blocking your collection.
If the index already exist, then the index-update time is so low that the time to add the document to the index is negligible compared to actually adding the document.
I have collection called TimeSheet having few thousands records now. This will eventually increase to 300 million records in a year. In this collection I embed few fields from another collection called Department which is mostly won't get any updates and only rarely some records will be updated. By rarely I mean only once or twice in a year and also not all records, only less than 1% of the records in the collection.
Mostly once a department is created there won't any update, even if there is an update, it will be done initially (when there are not many related records in TimeSheet)
Now if someone updates a department after a year, in a worst case scenario there are chances collection TimeSheet will have about 300 million records totally and about 5 million matching records for the department which gets updated. The update query condition will be on a index field.
Since this update is time consuming and creates locks, I'm wondering is there any better way to do it? One option that I'm thinking is run update query in batches by adding extra condition like UpdatedDateTime> somedate && UpdatedDateTime < somedate.
Other details:
A single document size could be about 3 or 4 KB
We have a replica set containing three replicas.
Is there any other better way to do this? What do you think about this kind of design? What do you think if there numbers I given are less like below?
1) 100 million total records and 100,000 matching records for the update query
2) 10 million total records and 10,000 matching records for the update query
3) 1 million total records and 1000 matching records for the update query
Note: The collection names department and timesheet, and their purpose are fictional, not the real collections but the statistics that I have given are true.
Let me give you a couple of hints based on my global knowledge and experience:
Use shorter field names
MongoDB stores the same key for each document. This repetition causes a increased disk space. This can have some performance issue on a very huge database like yours.
Pros:
Less size of the documents, so less disk space
More documennt to fit in RAM (more caching)
Size of the do indexes will be less in some scenario
Cons:
Less readable names
Optimize on index size
The lesser the index size is, the more it gets fit in RAM and less the index miss happens. Consider a SHA1 hash for git commits for example. A git commit is many times represented by first 5-6 characters. Then simply store the 5-6 characters instead of the all hash.
Understand padding factor
For updates happening in the document causing costly document move. This document move causing deleting the old document and updating it to a new empty location and updating the indexes which is costly.
We need to make sure the document don't move if some update happens. For each collection there is a padding factor involved which tells, during document insert, how much extra space to be allocated apart from the actual document size.
You can see the collection padding factor using:
db.collection.stats().paddingFactor
Add a padding manually
In your case you are pretty sure to start with a small document that will grow. Updating your document after while will cause multiple document moves. So better add a padding for the document. Unfortunately, there is no easy way to add a padding. We can do it by adding some random bytes to some key while doing insert and then delete that key in the next update query.
Finally, if you are sure that some keys will come to the documents in the future, then preallocate those keys with some default values so that further updates don't cause growth of document size causing document moves.
You can get details about the query causing document move:
db.system.profile.find({ moved: { $exists : true } })
Large number of collections VS large number of documents in few collection
Schema is something which depends on the application requirements. If there is a huge collection in which we query only latest N days of data, then we can optionally choose to have separate collection and old data can be safely archived. This will make sure that caching in RAM is done properly.
Every collection created incur a cost which is more than cost of creating collection. Each of the collection has a minimum size which is a few KBs + one index (8 KB). Every collection has a namespace associated, by default we have some 24K namespaces. For example, having a collection per User is a bad choice since it is not scalable. After some point Mongo won't allow us to create new collections of indexes.
Generally having many collections has no significant performance penalty. For example, we can choose to have one collection per month, if we know that we are always querying based on months.
Denormalization of data
Its always recommended to keep all the related data for a query or sequence of queries in the same disk location. You something need to duplicate the information across different documents. For example, in a blog post, you'll want to store post's comments within the post document.
Pros:
index size will be very less as number of index entries will be less
query will be very fast which includes fetching all necessary details
document size will be comparable to page size which means when we bring this data in RAM, most of the time we are not bringing other data along the page
document move will make sure that we are freeing a page, not a small tiny chunk in the page which may not be used in further inserts
Capped Collections
Capped collection behave like circular buffers. They are special type of fixed size collections. These collection can receive very high speed writes and sequential reads. Being fixed size, once the allocated space is filled, the new documents are written by deleting the older ones. However document updates are only allowed if the updated document fits the original document size (play with padding for more flexibility).
My application currently use MySQL. In order to support very fast deletion, I organize my data in partitions, according to timestamp. Then when data becomes obsolete, I just drop the whole partition.
It works great, and cleaning up my DB doesn't harm my application performance.
I would want to replace MySQL with MongoDB, and I'm wondering if there's something similiar in MongoDB, or would I just need to delete the records one by one (which, I'm afraid, will be really slow and will make my DB busy, and slow down queries response time).
In MongoDB, if your requirement is to delete data to limit the collection size, you should use a capped collection.
On the other hand, if your requirement is to delete data based on a timestamp, then a TTL index might be exactly what you're looking for.
From official doc regarding capped collections:
Capped collections automatically remove the oldest documents in the collection without requiring scripts or explicit remove operations.
And regarding TTL indexes:
Implemented as a special index type, TTL collections make it possible to store data in MongoDB and have the mongod automatically remove data after a specified period of time.
I thought, even though I am late and an answer has already been accepted, I would add a little more.
The problem with capped collections is that they regularly reside upon one shard in a cluster. Even though, in latter versions of MongoDB, capped collections are shardable they normally are not. Adding to this a capped collection MUST be allocated on the spot, so if you wish to have a long history before clearing the data you might find your collection uses up significantly more space than it should.
TTL is a good answer however it is not as fast as drop(). TTL is basically MongoDB doing the same thing, server-side, that you would do in your application of judging when a row is historical and deleting it. If done excessively it will have a detrimental effect on performance. Not only that but it isn't good at freeing up space to your $freelists which is key to stopping fragmentation in MongoDB.
drop()ing a collection will literally just "drop" the collection on the spot, instantly and gracefully giving that space back to MongoDB (not the OS) giving you absolutely no fragmentation what-so-ever. Not only that but the operation is a lot faster, 90% of the time, than most other alternatives.
So I would stick by my comment:
You could factor the data into time series collections based on how long it takes for data to become historical, then just drop() the collection
Edit
As #Zaid pointed out, even with the _id field capped collections are not shardable.
One solution to this is using TokuMX which supports partitioning:
https://www.percona.com/blog/2014/05/29/introducing-partitioned-collections-for-mongodb-applications/
Advantages over capped collections: capped collections use a fixed amount of space (even when you don't have this much data) and they can't be resized on-the-fly. Partitioned collections usage depends on data; you can add and remove partitions (for newly inserted data) as you see fit.
Advantages over TTL: TTL is slow, it just takes care of removing old data automatically. Partitions are fast - removing data is basically just a file removal.
HOWEVER: after getting acquired by Percona, development of TokuMX appears to have stopped (would love to be corrected on this point). Unfortunately MongoDB doesn't support this functionality and with TokuMX on its way out it looks like we will be stranded without proper solution.