In-Progress
| Jira Legacy | ||||||
|---|---|---|---|---|---|---|
|
Overview
FOLIO clients often have the need to call an API to look up a record by UUID, one at a time for thousands if not hundred of thousands of records. The obvious disadvantage of retrieving one record at a time is the overhead associates with it. Currently querying for one record incurs a SELECT count_estimate() call, which equals the actual SELECT query time. Additionally, each API call out to a storage module incurs a mod-authtoken call, which is needed to verify the token and permission of the caller. Therefore, this testing effort explores whether or not concatenating UUIDs in the CQL query string would be more efficient in retrieving records, and if there was any downside to doing this.
This test will
Summary
- Concatenating UUIDs in a batch can be done up to 50 records.
- Retrieving one record at a time at a rate of 40 requests/second uses up about 700MB of database RAM in 15 minutes. Retrieving more records at a time or having many concurrent users doing multi-records lookups concurrently will use up more DB memory and likely result the DB crashing when it runs out of memory.
- The optimal range in terms of time and resource is about 10-20 records for 1 user.
- It's up to the application developer to make a decision of concatenating UUIDs, and if so, how many UUIDs. Many applications have different needs, to look up a handful of records or hundreds of thousands of records. This report will lay out the facts for the developer to be informed about the pros and cons of concatenating UUIDs.
Test Results
The test that was run mimic the data-export workflow (PERF-98), making two calls: GET /inventory-storage/item-storage and GET /inventory-storage/holdings-storage but is not designed to behave in the same way as the actual data-exporting workflow, which would pause to process the retrieved records after each API call. More importantly UUIDs are concatenate in these calls which deviates from the real use case. These tests were run for 15 minutes each
...
With one user the request rate was able to be scaled down to a realistic 40 requests/sec. The average response time was less than 30ms for both API calls. Tests also ran for 40 and 50 UUIDs concatenation, and like the 8 users tests, request rate is cut down by half while the response time increased by half. The surprising thing was that even with 50 records the database did not crash during the test run.
CPU Utilization
This graph shows the CPU utilization of the modules in the first six tests. The series that we are interested in is mod-inventory-storage which has the highest of all the spikes. With the request rate that was thrown at it and the number of records that it has to deserialize from strings, the CPU utilization is quite high. Okapi CPU utilization is constant across the tests. mod-authtoken's CPU usage is rightly high because it's being called for each API call, but as expected its CPU utilization decreases when more records are concatenated as there is not as much need to go to mod-authentication. This decrease can be seen from 1 record to 10 records to 20 records to 30 and 40 records concatenation tests, with 40 UUIDs concatenation using the least CPU cycles.
...
The pattern seen in 8 virtual users tests with varied numbers of UUIDs concatenated are repeated here in the 1 virtual tests. Higher CPU utilizations for mod-circulation-storage as more UUIDs are being retrieved and processed, while mod-authtoken's CPU usage decreases.
Database
With 8 virtual users database CPU utilization across the tests doesn't vary much between 10 and 40 UUIDs concatenation tests. Retrieving 1 record at a time uses less CPU than when searching for many records at a time. Note the spikes and troughs in tests with 20, 30, and 40 records tests. This is due to memory issues, more below.
...
Why is the database using up so much memory? There were no slow queries, as the queries used in these lookup operations utilized indexes. The only plausible is that perhaps the count_estimate() calls used up and held up memories because its logic is fairly complex and might be resource intensive: https://github.com/folio-org/raml-module-builder#estimated-totalrecords
Discussions
Concatenating UUIDs seemed like a good idea, up to a point. It certainly returns results a faster due to less trips to mod-inventory-storage and less trips to mod-authtoken. However, the storage module works much harder to deserialize all the records coming back, and the database uses up memory at a much faster rate and could result in a DB crash minutes into the test run if looking up more than 20 records at once. The amount of load also causes issues. If only one user creates loads by the number of concatenated UUDIs then the database could handle it fairly easily, but if there are multiple users generating the same load then database errors would likely be generated. Likewise the more records concatenated result in more system resources being consumed, no matter the number of users. Therefore, it's up to the app developer to decide whether or not concatenating UUIDs in lookups make sense or not while taking into consideration the expected load and the number of concurrent users. This pattern does not apply for a use case that has many concurrent users, for example, but may be applicable for background tasks.
...