Raise default global-dict retained budget 1GB -> 2GB (#303 finding 2)#306
Merged
Conversation
) On a wide, high-cardinality file the per-column code buffers (~37 MB each, ADR 0021) sum past the 1 GB budget, and eviction demotes the largest-retained — i.e. the highest-cardinality — columns to per-chunk dictionaries, repeating their values pool every chunk. nyc-311 (~30 admitted string columns ≈ 1.15 GB retained) lost its three Cross Street / Street Name columns this way. Raise the default to 2 GB, which fits that file with headroom (measured: 2 GB and 3 GB both match the 8 GB result, so retained tops out ~1.15 GB) while still bounding the pathological many-wide-columns risk. Constrained-heap writers can still lower it via withGlobalDictMaxRetainedBytes(...). nyc-311 re-encode: 1934.23 MB -> 1879.34 MB (1.10x -> 1.07x vortex-jni). This overlaps #305 (both improve the street columns, via keeping-global vs bitpacked codes): stacked on #305 the budget adds ~35 MB net (1862 -> 1827 MB). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
On a wide, high-cardinality file, the global-dict retained budget (1 GB) is too tight: the writer buffers each admitted column's per-chunk codes in memory until
close()(~37 MB/column, the ADR 0021 design), and with ~30 dict columns on nyc-311 the total (~1.15 GB) crosses the budget. Eviction then demotes the largest-retained columns — the highest-cardinality ones — to per-chunk dictionaries, repeating their values pool each chunk. That's exactly what happened to Cross Street 1/2 and Street Name (#303 finding 2).How
Raise the default
globalDictMaxRetainedBytesfrom 1 GB to 2 GB. Measured: 2 GB, 3 GB, and 8 GB all produce the identical file, so nyc-311's retained total tops out ~1.15 GB and 2 GB fits it with headroom while still bounding the pathological many-wide-columns risk. Constrained-heap writers can lower it viawithGlobalDictMaxRetainedBytes(...).Tradeoff (why this is a policy bump, not a pure bug fix)
The budget exists to cap writer memory. Raising it lets the writer buffer up to 2 GB before evicting, so a wide/tall enough file uses more heap. This is a deliberate size↔memory tradeoff, kept overridable per-write.
Impact (real nyc-311 corpus, off
main)−54.9 MB off main. Note this overlaps #305 (both improve the same street columns — this PR keeps them global, #305 bitpacks per-chunk codes): stacked on #305 the budget adds ~35 MB net (1862 → 1827 MB), not another 55.
Testing
WriteOptionsTestdefault assertions updated to 2 GB../mvnw verifygreen.Closes #303.