What are Partitions?
Partitions route your data to an index becoming a separate subset of data in your account. Creating smaller and separate subsets of data is central to search optimization. When you run a search against an index, results are returned more quickly and efficiently because the search runs against a smaller data set.
This example shows a customer that created three additional Partitions to separate data by environment.
Consider the following queries:
|Query 2||Partitions in place|
|Query 3||Partitions in place|
|Query 4||Partitions in place|
- Query 1: There are no custom Partitions created and you only have the Default Index, 100% of your data is scanned in order to find all production log messages for the Snort security app.
- Query 2: Partitions do exist,
_index=prodlimits the scope of the query and only about 40% of the data is scanned to get the same results as Query 1. But it is redundant.
- Query 3: You can take advantage of Partitions without having to rewrite your existing queries. Sumo Logic's behind-the-scenes Query Rewriting, performed for queries run against data in the Continuous Tier, is smart enough to understand that the scope of what you are looking for is included within
_index=prod; therefore at runtime, it will rewrite the query as Query 2.
- Query 4: We want to search for data that is in a custom Partition, as well as data that exists in the Default Index. However, query rewriting does not have the ability to OR indexes together. Instead, another behind-the-scenes feature, Inverse View Rewriting kicks in, we know that the data is NOT contained in the DEV and QA index, so those will be skipped. This query will only scan the Prod index and the Default Index.
What is Query Rewriting?
Whenever possible, we rewrite a user's queries to perform better. In the following example it will take advantage of existing Partitions to rewrite queries.
This means that:
Will be rewritten as:
_index=prod AND _sourceCategory=prod/security/snort
This is possible because:
The example environment is using a robust
The Partition was scoped using
The searches are using
_sourceCategory, so they can easily be mapped to Partitions
The scope of this search (
_sourceCategory=prod/security/snort) falls within the scope of the Partition (
Therefore, defining a broad scope for your Partitions (for example,
_sourceCategory=prod/*), and searching with
_sourceCategory allows you to take advantage of query rewriting, and it allows you to potentially not have to manually rewrite your existing queries.
Query rewriting does not affect the volume of data scanned when you run a query in the Infrequent Data Tier, so it has no impact on the cost of Infrequent queries.
Create a Partition
As an Admin, you create Partitions by specifying their routing expression. We recommend you use
_sourceCategeory to define your routing expressions to take full advantage of Query Rewriting.
The following example shows the routing expression for the three custom Partitions:
Here are simple steps to create the Dev Partition:
How can my team use Partitions?
Once created, Partitions can be used by anyone in your account, helping you reduce the scope of your searches and improve the performance for all users. Query 2 above takes advantage of our newly created Partition to scan only 40% of the data. As noted above, Query 3 is also a good option, because Query Rewriting will produce the same results as Query 1. This might eliminate the need to edit all your queries once your Partitions are in place.
Here's an example of a search using the Prod Partition to narrow the search scope:
Best Practices when using Partitions
Avoid creating too many partitions to avoid fragmentation.
We recommend 20 as the maximum number of partitions. This is both to avoid index fragmentation and data management issues. This includes communicating to users which partition to use and duplication across partitions.
Optimal partitions are sized between 1% and 30% of total ingest.
Partitions that are too small may cause index fragmentation and degraded search performance. It is possible to create partitions larger than 30% without adverse effects, however the performance gains will be diminished.
Don’t create overlapping partitions.
This will lead to duplication of data (increasing your billed ingest rate), and degraded performance. Sumo Logic will not return duplicate results, but the process of de-duplication is time consuming and will increase query durations.
Do not use the NOT operator in partition definitions.
This will likely exclude data that should be contained within your partition and will reduce the chances that your partition will be reused by queries that are rewritten.
Do not use sourceHost to define your partitions.
It may prevent you from searching horizontally without OR’ing partitions together.
Use an intuitive naming scheme
This helps users easily identify the correct partition to use.
Keep your partition broadly scoped with sourceCategory and avoid keywords.
Use sourceCategory in your partitions definitions and avoid keywords to keep your partition broadly scoped. You can always narrow down the scope of your search when you query your partition.
Group similar data together.
In the example above, we used prod/QA/Dev environment, as you will most often be searching across all your Prod data. If you need to search across environments, you can OR 2 or more Partitions.
Partitions are great, where can I find more information?
See Manage Partitions.