On this put up I discover methods to help analytical queries with out encountering prohibitive scan prices, by leveraging secondary indexes in DynamoDB. I additionally consider the professionals and cons of this strategy in distinction to extracting knowledge to a different system like Athena, Spark or Elastic.
Rockset not too long ago added help for DynamoDB – which mainly means you may run quick SQL on DynamoDB tables with none ETL. As I spoke to our customers, I got here throughout other ways by which world secondary indexes (GSI) are used for analytical queries.
DynamoDB shops knowledge beneath the hood by partitioning it over a lot of nodes primarily based on a user-specified partition key subject current in every merchandise. This user-specified partition key might be optionally mixed with a kind key to characterize a major key. The first key acts as an index, making question operations on it cheap. A question operation can do equality comparability (=) on the partition key and comparative operations (>, <, =, BETWEEN) on the kind key if specified. Performing operations that aren’t lined by the above scheme requires using a scan operation, which is often executed by scanning over your complete DynamoDB desk in parallel. These scans might be gradual and costly by way of Learn Capability Models (RCUs) as a result of they require a full learn of your complete desk. Scans additionally are likely to decelerate when the desk measurement grows as there may be extra knowledge to scan to supply outcomes.
If we need to help analytical queries with out encountering prohibitive scan prices, we are able to leverage secondary indexes in DynamoDB. Secondary indexes additionally consist of making partition keys and elective kind keys over fields that we need to question over in a lot the identical manner as the first key. Secondary indexes are sometimes used to enhance utility efficiency by indexing fields that are queried fairly often. Question operations on secondary indexes can be used to energy particular options via analytic queries which have clearly outlined necessities—like computing a leaderboard in a sport. One clear benefit of this strategy of performing analytical queries is that there isn’t any want for another system.
Nonetheless, it’s infeasible to make use of this strategy for a wider vary of analytical queries due to the restricted forms of queries it helps. The total gamut of analytics requires filtering on a number of fields, grouping, ordering, becoming a member of knowledge between knowledge units, and many others., which can’t be achieved merely via secondary indexes. Secondary indexes that may be created are additionally restricted in quantity and require some planning to make sure that they scale nicely with the information. A badly chosen partition key can worsen efficiency and improve prices considerably. Knowledge in DynamoDB can have a nested construction together with arrays and objects, however indexes can solely be constructed on sure primitive sorts. This could pressure denormalizing of the information to flatten nested objects and arrays with a purpose to construct secondary indexes, which might probably explode the variety of writes carried out and related prices. Aside from price and suppleness, there are additionally safety and efficiency issues relating to supporting analytic use circumstances on an operational knowledge retailer in a manufacturing setting.
Benefits
- No further setup exterior DynamoDB
- Quick and scalable serving for primary analytical queries over listed fields
Disadvantages
- Costly when queries require scans over DynamoDB
- Very restricted help for analytical queries over indexes; no SQL queries, grouping, or joins
- Can’t arrange indexes on nested fields with out denormalizing knowledge and exploding out writes
- Safety and efficiency implications of working analytical queries on an operational database
This strategy could also be appropriate if we’ve an utility that requires a particular characteristic that’s easy sufficient to be realized utilizing a question over an index. The elevated storage and I/O price and the restricted question potential make it unsuitable for the broader vary of analytical queries in any other case. Subsequently, for a majority of analytic use circumstances, it’s price efficient to export the information from DynamoDB into a unique system that enables us to question with increased constancy.
If you’re contemplating extracting knowledge to a different system, there are a number of completely different choices for real-time analytics:
- DynamoDB + Glue + S3 + Athena
- DynamoDB + Hive/Spark
- DynamoDB + AWS Lambda + Elasticsearch
- DynamoDB + Rockset
I evaluate every of those by way of ease of setup, upkeep, question functionality, latency in my different weblog put up Analytics on DynamoDB: Evaluating Athena, Spark and Elastic, the place I additionally consider which use circumstances every of them are finest suited to.
Different DynamoDB sources: