vamsikarnika opened a new pull request, #18818:
URL: https://github.com/apache/hudi/pull/18818
CompressionHandler currently allocates a new Deflater on every
compressBytes() and a new Inflater on every decompressBytes(). On JDK 8 both
classes register a Finalizer on construction. Under sustained, multi-threaded
disk-map traffic (observed during MDT/RLI compaction merging millions of
records across several Spark task threads on the same executor), the rate of
zlib allocations exceeds the rate at which the single Finalizer thread can
drain its queue. Native ZStreamRef handles pile up in old gen, the heap
saturates, and G1 enters a mixed-GC death spiral while application threads make
no progress.
CompressionHandler is already held in a ThreadLocal, so a single Deflater/
Inflater pair per worker thread is sufficient. This change:
- Adds transient Deflater/Inflater fields and lazy accessors (transient so
the class remains Serializable; lazy so deserialized instances rebuild the
codecs on first use).
- Calls reset() on the cached codecs at the start of each call.
- Passes the user-supplied codecs to DeflaterOutputStream(out, def) and
InflaterInputStream(in, inf), which sets usesDefaultDeflater/Inflater to false
so close() does not call end() on the codec — the codec survives the
try-with-resources for reuse on the next call.
On-disk format, compression level, and error semantics are unchanged.
Allocation rate drops from O(records) to O(threads). On JDK 9+ this also
removes per-call Cleaner registration overhead.
### Describe the issue this Pull Request addresses
<!-- Either describe the issue inline here with motivation behind the
changes
(or) link to an issue by including `Closes #<issue-number>` for
context.
If this PR includes changes to the storage format, public APIs,
or has breaking changes, use `!` (e.g., feat!: ...) -->
### Summary and Changelog
<!-- Short, plain-English summary of what users gain or what changed in
behavior.
Followed by a detailed log of all the changes. Highlight if any code
was copied. -->
### Impact
<!-- Describe any public API or user-facing feature change or any
performance impact. -->
### Risk Level
<!-- Accepted values: none, low, medium or high. Other than `none`, explain
the risk.
If medium or high, explain what verification was done to mitigate the
risks. -->
### Documentation Update
<!-- Describe any necessary documentation update if there is any new
feature, config, or user-facing change. If not, put "none".
- The config description must be updated if new configs are added or the
default value of the configs are changed.
- Any new feature or user-facing change requires updating the Hudi website.
Please follow the
[instruction](https://hudi.apache.org/contribute/developer-setup#website)
to make changes to the website. -->
### Contributor's checklist
- [ ] Read through [contributor's
guide](https://hudi.apache.org/contribute/how-to-contribute)
- [ ] Enough context is provided in the sections above
- [ ] Adequate tests were added if applicable
--
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]
