Introduce JK to Disable Memory Reallocation Optimization #325
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
meta-cla
bot
added
the
CLA Signed
|
@macvincent has exported this pull request. If you are a Meta employee, you can view the originating Diff in D87494427. |
macvincent
added a commit
to macvincent/nimble
that referenced
this pull request
Nov 20, 2025
…ubator#325) Summary: When implementing the stream chunker, we anticipated that stream buffers after chunking will end up growing to the size that previously triggered chunking. As a tradeoff between minimizing reallocations (for performance) and actually releasing memory (to relieve memory pressure), we heuristically determine the new buffer capacity for each stream to be larger that required. The issue with this optimization is that it conflicts with the rest of our memory tracking logic since we now have retained memory in the memory pool that is not accounted for. We now know through local testing that disabling this optimization leads to better memory pressure relief. In this diff, we introduce a JK, [dwio/nimble_chunking:disable_memory_reallocation_optimization](https://www.internalfb.com/intern/justknobs/?name=dwio%2Fnimble_chunking#disable_memory_reallocation_optimization), that will be enabled just for DISCO experiments. This will help us understand the full impact of this optimization and whether it should be retained. Differential Revision: D87494427
macvincent
force-pushed
the
export-D87494427
branch
from
1e42244 to
69ff82f
Compare
…ubator#325) Summary: When implementing the stream chunker, we anticipated that stream buffers after chunking will end up growing to the size that previously triggered chunking. As a tradeoff between minimizing reallocations (for performance) and actually releasing memory (to relieve memory pressure), we heuristically determine the new buffer capacity for each stream to be larger that required. The issue with this optimization is that it conflicts with the rest of our memory tracking logic since we now have retained memory in the memory pool that is not accounted for. We now know through local testing that disabling this optimization leads to better memory pressure relief. In this diff, we introduce a JK, [dwio/nimble_chunking:disable_memory_reallocation_optimization](https://www.internalfb.com/intern/justknobs/?name=dwio%2Fnimble_chunking#disable_memory_reallocation_optimization), that will be enabled just for DISCO experiments. This will help us understand the full impact of this optimization and whether it should be retained. Differential Revision: D87494427
macvincent
force-pushed
the
export-D87494427
branch
from
69ff82f to
07208de
Compare
macvincent
added a commit
to macvincent/nimble
that referenced
this pull request
Nov 20, 2025
…ubator#325) Summary: When implementing the stream chunker, we anticipated that stream buffers after chunking will end up growing to the size that previously triggered chunking. As a tradeoff between minimizing reallocations (for performance) and actually releasing memory (to relieve memory pressure), we heuristically determine the new buffer capacity for each stream to be larger that required. The issue with this optimization is that it conflicts with the rest of our memory tracking logic since we now have retained memory in the memory pool that is not accounted for. We now know through local testing that disabling this optimization leads to better memory pressure relief. We performed local DISCO tests with and without this optimization for two tables and included the comparison in https://docs.google.com/spreadsheets/d/1kZvBwhVHZRyB7tg-qT2Et4V_7mDWyNr-VrtDzjG_FKY/view?gid=1209014630#gid=1209014630. It shows that for these two tables we save an average of **15%** improvement in average write memory. We also see **4%** write CPU improvement on average. Though we saw **-6%** regression in CPU for a single 256MB stripe experiment. These results indicate that more testing is required in order to understand its impact on a larger sample size. In this diff, we introduce a JK, [dwio/nimble_chunking:disable_memory_reallocation_optimization](https://www.internalfb.com/intern/justknobs/?name=dwio%2Fnimble_chunking#disable_memory_reallocation_optimization), that will be enabled just for DISCO experiments. This will help us understand the full impact of this optimization and whether it should be retained. Differential Revision: D87494427
macvincent
added a commit
to macvincent/nimble
that referenced
this pull request
Nov 20, 2025
…ubator#325) Summary: When implementing the stream chunker, we anticipated that stream buffers after chunking will end up growing to the size that previously triggered chunking. As a tradeoff between minimizing reallocations (for performance) and actually releasing memory (to relieve memory pressure), we heuristically determine the new buffer capacity for each stream to be larger that required. The issue with this optimization is that it conflicts with the rest of our memory tracking logic since we now have retained memory in the memory pool that is not accounted for. We now know through local testing that disabling this optimization leads to better memory pressure relief. We performed local DISCO tests with and without this optimization for two tables and included the comparison in https://docs.google.com/spreadsheets/d/1kZvBwhVHZRyB7tg-qT2Et4V_7mDWyNr-VrtDzjG_FKY/view?gid=1209014630#gid=1209014630. It shows that for these two tables we save an average of **15%** improvement in average write memory. We also see **4%** write CPU improvement on average. Though we saw **-6%** regression in CPU for a single 256MB stripe experiment. These results indicate that more testing is required in order to understand its impact on a larger sample size. In this diff, we introduce a JK, [dwio/nimble_chunking:disable_memory_reallocation_optimization](https://www.internalfb.com/intern/justknobs/?name=dwio%2Fnimble_chunking#disable_memory_reallocation_optimization), that will be enabled just for DISCO experiments. This will help us understand the full impact of this optimization and whether it should be retained. Differential Revision: D87494427
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary: When implementing the stream chunker, we anticipated that stream buffers after chunking will end up growing to the size that previously triggered chunking. As a tradeoff between minimizing reallocations (for performance) and actually releasing memory (to relieve memory pressure), we heuristically determine the new buffer capacity for each stream. The issue with this optimization is that it conflicts with the rest of our memory tracking logic since we now have retained memory in the memory pool that is not accounted for. We now through local testing that disabling this optimization leads to better memory pressure relief. In this diff, we introduce a JK, dwio/nimble_chunking:disable_memory_reallocation_optimization, that will be enabled just for DISCO experiments. This will help us understand the full impact of this optimization and whether it should be retained.
Differential Revision: D87494427