[Feature](iceberg) Add manifest-level cache for Iceberg tables to reduce I/O and parsing overhead#59056
Merged
morningman merged 8 commits intoapache:masterfrom Dec 26, 2025
Merged
Conversation
Contributor
|
Thank you for your contribution to Apache Doris. Please clearly describe your PR:
|
Contributor
Author
|
run buildall |
TPC-H: Total hot run time: 36570 ms |
TPC-DS: Total hot run time: 180705 ms |
ClickBench: Total hot run time: 27.21 s |
Contributor
Author
|
run buildall |
suxiaogang223
changed the title
TPC-H: Total hot run time: 36120 ms |
TPC-DS: Total hot run time: 178998 ms |
ClickBench: Total hot run time: 27.32 s |
Contributor
FE UT Coverage ReportIncrement line coverage |
Contributor
FE Regression Coverage ReportIncrement line coverage |
Contributor
Author
|
run buildall |
TPC-H: Total hot run time: 35151 ms |
Contributor
Author
|
run buildall |
TPC-H: Total hot run time: 35385 ms |
TPC-DS: Total hot run time: 178544 ms |
ClickBench: Total hot run time: 27.24 s |
Contributor
FE Regression Coverage ReportIncrement line coverage |
suxiaogang223
force-pushed
the
manifest_cache2
branch
from
5c598d1 to
53318e7
Compare
Contributor
Author
|
run buildall |
TPC-H: Total hot run time: 34962 ms |
TPC-DS: Total hot run time: 178232 ms |
ClickBench: Total hot run time: 27.41 s |
Contributor
FE Regression Coverage ReportIncrement line coverage |
Contributor
Author
|
run buildall |
TPC-H: Total hot run time: 35242 ms |
TPC-DS: Total hot run time: 178687 ms |
Contributor
Author
|
run external |
… remove obsolete explain test
Contributor
Author
|
run buildall |
TPC-H: Total hot run time: 36088 ms |
TPC-DS: Total hot run time: 179920 ms |
ClickBench: Total hot run time: 27.17 s |
Contributor
FE UT Coverage ReportIncrement line coverage |
Contributor
FE Regression Coverage ReportIncrement line coverage |
morningman
approved these changes
Dec 26, 2025
Contributor
|
PR approved by at least one committer and no changes requested. |
github-actions
bot
added
the
approved
Contributor
|
PR approved by anyone and no changes requested. |
CalvinKirs
approved these changes
Dec 26, 2025
github-actions bot
pushed a commit
that referenced
this pull request
Dec 26, 2025
…uce I/O and parsing overhead (#59056) ### What problem does this PR solve? ## Motivation During Iceberg query planning, FE needs to read and parse the metadata chain: ManifestList → Manifest → DataFile/DeleteFile. When frequently querying hot partitions or executing small batch queries, the same Manifest files are repeatedly read and parsed, causing significant I/O and CPU overhead. ## Solution This PR introduces a manifest-level cache (`IcebergManifestCache`) in FE to cache the parsed DataFile/DeleteFile lists per manifest file. The cache is implemented using Caffeine with weight-based LRU eviction and TTL support. ### Key Components - **IcebergManifestCache**: Core cache implementation using Caffeine - Weight-based LRU eviction controlled by `iceberg.manifest.cache.capacity-mb` - TTL expiration via `iceberg.manifest.cache.ttl-second` - Single-flight loading to prevent duplicate parsing of the same manifest - **ManifestCacheKey**: Cache key consisting of: - Manifest file path - **ManifestCacheValue**: Cached payload containing: - List of `DataFile` or `DeleteFile` - Estimated memory weight for eviction - **IcebergManifestCacheLoader**: Helper class to load and populate the cache using `ManifestFiles.read()` ### Cache Invalidation Strategy - Key changes automatically invalidate stale entries (length/lastModified/sequenceNumber changes) - TTL prevents stale data when underlying storage doesn't support precise mtime/etag - Different snapshots use different manifest paths/keys, ensuring snapshot-level isolation ### Iceberg Catalog Properties | Config | Default | Description | |--------|---------|-------------| | `iceberg.manifest.cache.enable` | `true` | Enable/disable manifest cache | | `iceberg.manifest.cache.capacity-mb` | `1024` | Maximum cache capacity in MB | | `iceberg.manifest.cache.ttl-second` | `48 * 60 * 60` | Cache entry expiration after access | ### Integration Point The cache is integrated in `IcebergScanNode.planFileScanTaskWithManifestCache()`, which: 1. Loads delete manifests via cache and builds `DeleteFileIndex` 2. Loads data manifests via cache and creates `FileScanTask` for each data file 3. Falls back to original scan if cache loading fails
8 tasks
16 tasks
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
7 participants
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 problem does this PR solve?
Motivation
During Iceberg query planning, FE needs to read and parse the metadata chain: ManifestList → Manifest → DataFile/DeleteFile. When frequently querying hot partitions or executing small batch queries, the same Manifest files are repeatedly read and parsed, causing significant I/O and CPU overhead.
Solution
This PR introduces a manifest-level cache (
IcebergManifestCache) in FE to cache the parsed DataFile/DeleteFile lists per manifest file. The cache is implemented using Caffeine with weight-based LRU eviction and TTL support.Key Components
IcebergManifestCache: Core cache implementation using Caffeine
iceberg.manifest.cache.capacity-mbiceberg.manifest.cache.ttl-secondManifestCacheKey: Cache key consisting of:
ManifestCacheValue: Cached payload containing:
DataFileorDeleteFileIcebergManifestCacheLoader: Helper class to load and populate the cache using
ManifestFiles.read()Cache Invalidation Strategy
Iceberg Catalog Properties
iceberg.manifest.cache.enabletrueiceberg.manifest.cache.capacity-mb1024iceberg.manifest.cache.ttl-second48 * 60 * 60Integration Point
The cache is integrated in
IcebergScanNode.planFileScanTaskWithManifestCache(), which:DeleteFileIndexFileScanTaskfor each data fileRelease note
None
Check List (For Author)
Test
Behavior changed:
Does this need documentation?
Check List (For Reviewer who merge this PR)