[Proxy] Proxy uses a lot of heap memory when uploading large function jar files #10908
Labels
type/bug
The PR fixed a bug or issue reported a bug
Comments
|
@addisonj would you like to pick this one? |
sijie
pushed a commit
that referenced
this issue
Jun 22, 2021
Fixes #10908 ### Motivation Pulsar Proxy uses a lot of heap memory when uploading large function jar files. This also leads to high GC activity since a continuous block of memory (byte array for the size of the upload) is allocated. GC will have to do compaction for the heap (which gets fragmented) to find a continuous block of memory. This is the reason why allocating large arrays are costly from GC perspective. The buffering solution added as part of #5361. The solution buffers also very large uploads to memory. ### Modifications * Limit the replay buffer size to a configurable limit which defaults to 5MB. This is configured with the `httpInputMaxReplayBufferSize` proxy configuration parameter. * Add unit test to see that buffer size gets limited * Add unit test for #5361
eolivelli
pushed a commit
to datastax/pulsar
that referenced
this issue
Jun 23, 2021
Fixes apache#10908 ### Motivation Pulsar Proxy uses a lot of heap memory when uploading large function jar files. This also leads to high GC activity since a continuous block of memory (byte array for the size of the upload) is allocated. GC will have to do compaction for the heap (which gets fragmented) to find a continuous block of memory. This is the reason why allocating large arrays are costly from GC perspective. The buffering solution added as part of apache#5361. The solution buffers also very large uploads to memory. ### Modifications * Limit the replay buffer size to a configurable limit which defaults to 5MB. This is configured with the `httpInputMaxReplayBufferSize` proxy configuration parameter. * Add unit test to see that buffer size gets limited * Add unit test for apache#5361 (cherry picked from commit 2324618)
yangl
pushed a commit
to yangl/pulsar
that referenced
this issue
Jun 23, 2021
Fixes apache#10908 ### Motivation Pulsar Proxy uses a lot of heap memory when uploading large function jar files. This also leads to high GC activity since a continuous block of memory (byte array for the size of the upload) is allocated. GC will have to do compaction for the heap (which gets fragmented) to find a continuous block of memory. This is the reason why allocating large arrays are costly from GC perspective. The buffering solution added as part of apache#5361. The solution buffers also very large uploads to memory. ### Modifications * Limit the replay buffer size to a configurable limit which defaults to 5MB. This is configured with the `httpInputMaxReplayBufferSize` proxy configuration parameter. * Add unit test to see that buffer size gets limited * Add unit test for apache#5361
kaushik-develop
pushed a commit
to kaushik-develop/pulsar
that referenced
this issue
Jun 24, 2021
Fixes apache#10908 ### Motivation Pulsar Proxy uses a lot of heap memory when uploading large function jar files. This also leads to high GC activity since a continuous block of memory (byte array for the size of the upload) is allocated. GC will have to do compaction for the heap (which gets fragmented) to find a continuous block of memory. This is the reason why allocating large arrays are costly from GC perspective. The buffering solution added as part of apache#5361. The solution buffers also very large uploads to memory. ### Modifications * Limit the replay buffer size to a configurable limit which defaults to 5MB. This is configured with the `httpInputMaxReplayBufferSize` proxy configuration parameter. * Add unit test to see that buffer size gets limited * Add unit test for apache#5361
nicoloboschi
pushed a commit
to datastax/pulsar
that referenced
this issue
Feb 28, 2022
Fixes apache#10908 Pulsar Proxy uses a lot of heap memory when uploading large function jar files. This also leads to high GC activity since a continuous block of memory (byte array for the size of the upload) is allocated. GC will have to do compaction for the heap (which gets fragmented) to find a continuous block of memory. This is the reason why allocating large arrays are costly from GC perspective. The buffering solution added as part of apache#5361. The solution buffers also very large uploads to memory. * Limit the replay buffer size to a configurable limit which defaults to 5MB. This is configured with the `httpInputMaxReplayBufferSize` proxy configuration parameter. * Add unit test to see that buffer size gets limited * Add unit test for apache#5361 (cherry picked from commit 2324618) (cherry picked from commit 7fa88cc)
bharanic-dev
pushed a commit
to bharanic-dev/pulsar
that referenced
this issue
Mar 18, 2022
Fixes apache#10908 ### Motivation Pulsar Proxy uses a lot of heap memory when uploading large function jar files. This also leads to high GC activity since a continuous block of memory (byte array for the size of the upload) is allocated. GC will have to do compaction for the heap (which gets fragmented) to find a continuous block of memory. This is the reason why allocating large arrays are costly from GC perspective. The buffering solution added as part of apache#5361. The solution buffers also very large uploads to memory. ### Modifications * Limit the replay buffer size to a configurable limit which defaults to 5MB. This is configured with the `httpInputMaxReplayBufferSize` proxy configuration parameter. * Add unit test to see that buffer size gets limited * Add unit test for apache#5361
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Describe the bug
Proxy uses a lot of heap memory when uploading large function jar files. This is due to the buffering solution added as part of #5361 . The solution buffers also very large uploads to memory.
To Reproduce
Create a function with a large jar file to reproduce.
The repro for #10906 , https://github.com/lhotari/pulsar-playground/tree/master/proxy-tls-issue-repro-2.7.2 , reproduces this issue too.
Expected behavior
The proxy heap memory usage shouldn't grow out of bounds. File uploads shouldn't be buffered in heap memory.
The text was updated successfully, but these errors were encountered: