Caideyipi opened a new pull request, #18851:
URL: https://github.com/apache/iotdb/pull/18851

   ## Description
   
   Several collection materialization paths already know how many entries they 
will append, but allocate default-sized containers. Large batches repeatedly 
grow and copy `ArrayList` backing arrays or resize hash tables.
   
   Pre-size 132 collections in 74 production Java files across eight modules:
   
   - Schema/path/plan deserialization, Pipe file-seal and handshake payloads, 
and progress metadata.
   - Session batch-write buffers and schema lists, and JDBC result column 
metadata.
   - Query projections, expression transformers, row-pattern evaluation, and 
spill readers.
   - Memtable tablet schema lists and recovery, TsFile device sets, ConfigNode 
plans/procedures, subscription retention snapshots, and UDF pattern processing.
   
   Use serialized entry counts or the exact number of outputs produced from the 
input. Keep optional empty lists at capacity zero and preserve their existing 
null/empty handling. Hash maps and sets use the existing Guava expected-size 
factories, which account for the default load factor and avoid capacity 
arithmetic overflow.
   
   This is independent of #18845; its WAL metadata and subscription 
acknowledgement changes are not included.
   
   ## Validation
   
   - Spotless applied; Checkstyle, Spotless, and forbidden API checks passed 
during the reactor build.
   - English reactor `test-compile` passed for the eight affected modules and 
their dependencies (30 modules).
   - 47 existing test classes passed: 486 tests, zero failures/errors/skips. 
Coverage includes Session/JDBC, schema filters and paths, plan/procedure 
serialization, memtable partial inserts and recovery, expression/operator 
behavior, TsFile metadata, Pipe payloads, and UDF pattern matching.
   - Chinese locale reactor `test-compile` passed for the same 30 modules with 
`-P with-zh-locale`.
   
   A separate JDK 17 allocation experiment used shared, pre-boxed integer 
entries and `ThreadMXBean.getThreadAllocatedBytes` after warm-up. For 
constructing and filling 1,024 entries:
   
   | Container | Default capacity | Pre-sized | Allocation reduction |
   | --- | ---: | ---: | ---: |
   | ArrayList | 15,024 bytes | 4,136 bytes | 72.5% |
   | HashMap | 49,264 bytes | 41,024 bytes | 16.7% |
   | HashSet | 49,280 bytes | 41,040 bytes | 16.7% |
   
   These are container allocation measurements; cluster throughput has not been 
benchmarked.
   
   - [x] Self-reviewed.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to