[ColinLeeo]

Motivation

TsFileIOReader::get_timeseries_indexes() resolved every requested measurement on its own: a binary search into the device's measurement index, one pread, a deserialize into a scratch PageArena, and a clone_from() into the caller's arena. A wide table pays that path once per column, and do_load_all_timeseries_index() also issued one pread per measurement. Loading the indexes of a 4500-column device therefore dominated the fixed cost of a table query.

Changes

• For projections of 16 or more measurements, get_timeseries_indexes() now walks the measurement index once (get_all_leaf()), then loads the device's whole TimeseriesIndex range and maps the results by measurement name. A narrow projection keeps the per-measurement path, where a full walk would deserialize every column to find a handful.
• do_load_all_timeseries_index() now reads the whole [first start, last end) range with one pread when the leaf entries tile it (they do for every normal file; a per-entry fallback remains for the rest) and deserializes each index straight into the given arena. This also speeds up the prepared-series (alloc_multi_ssi) and device-metadata paths that share the helper.

Nothing about the returned indexes changes: same chunk metadata, same aligned wrapper (AlignedTimeseriesIndex with the shared time index), same nullptr for a measurement the device does not have.

Tests

New cpp/test/file/timeseries_index_batch_load_test.cc (40 measurements, non-aligned and aligned devices):

• LoadsEveryMeasurementOfWideDevice — bulk path returns every column with its chunk metadata.
• MissingMeasurementBecomesNull — an absent name in a bulk request stays nullptr.
• AlignedDeviceSharesTheTimeIndex — bulk path wraps aligned columns with the device's time index.
• NarrowRequestUsesPerMeasurementPath — the per-measurement path still resolves below the threshold.
• MatchesPerMeasurementPath — chunk offsets of the bulk path equal those of the per-measurement path for the same columns.

Full TsFile_Test suite passes; mvn -P with-cpp spotless:check passes.

Benchmark

get_timeseries_indexes() for every measurement of one device (2 rows per series, release build, macOS, 10 devices per iteration, fresh PageArena per query):

measurements	develop	this PR
4500	41.5–46.5 ms/device	0.53 ms/device
40	0.074 ms/device	0.006 ms/device
8	0.007 ms/device	0.007 ms/device (per-measurement path)

The 4500-column case is index loading only, not a full query; for a wide-table scan it removes the per-column arena/deserialize/clone cost, which was a large part of the device's fixed read overhead.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 68.96552% with 18 lines in your changes missing coverage. Please review.
✅ Project coverage is 66.55%. Comparing base (90c7862) to head (2afbd45).
⚠️ Report is 2 commits behind head on develop.

Files with missing lines	Patch %	Lines
cpp/src/file/tsfile_io_reader.cc	68.96%	4 Missing and 14 partials ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1095      +/-   ##
===========================================
+ Coverage    66.51%   66.55%   +0.03%     
===========================================
  Files          778      782       +4     
  Lines        56577    56743     +166     
  Branches      9105     9149      +44     
===========================================
+ Hits         37633    37764     +131     
- Misses       17291    17308      +17     
- Partials      1653     1671      +18     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Warning

Copilot couldn't run its full agentic review because it didn't start before the timeout. Make sure your repository has a runner available, or add a copilot-code-review.yml file specifying one with the runs-on attribute. See the docs for more details.

Copilot review overview

5 open issues

• Allocating the entire read buffer (ti_buf) from PageArena means the raw bytes for the whole… · New
• This test allocates MeasurementSchema with new and passes raw pointers, but it's not clear from… · New
• Asserting chunks->size() == 1 can make this test brittle if chunking behavior changes (e.g.,… · New
• The comment says it deserializes \"every TimeseriesIndex\" in the range, but the implementation… · New
• The comment is inaccurate: GoogleTest typically runs all tests in a single process by default.… · New

What changed in this PR

Improves C++ TsFile timeseries index lookup performance for wide projections by batch-loading timeseries index blocks and reducing per-measurement syscalls/deserialization overhead.

Changes:

• Add a batch-load fast path in TsFileIOReader::get_timeseries_indexes() for requests ≥16 measurements.
• Optimize do_load_all_timeseries_index() to pread a contiguous [first,last) range in one call when leaf entries tile the range, with fallback to per-entry reads.
• Add a focused C++ test suite validating correctness for batch-load behavior (missing measurements, aligned devices, and parity with per-measurement path).

File	Description
cpp/​test/​file/​timeseries_index_batch_load_test.cc	Adds unit tests covering batch-load behavior, aligned wrapping, missing measurements, and consistency with the old per-measurement loader.
cpp/​src/​file/​tsfile_io_reader.h	Declares the new helper API for loading a contiguous TimeseriesIndex byte range.
cpp/​src/​file/​tsfile_io_reader.cc	Implements batch-load logic, range-loading helper, and single-pread optimization when leaf entries tile a contiguous region.

_{Review effort: Lite}

[Copilot]

Warning

Copilot couldn't run its full agentic review because it didn't start before the timeout. Make sure your repository has a runner available, or add a copilot-code-review.yml file specifying one with the runs-on attribute. See the docs for more details.

Copilot review overview

5 open issues

• Batch-loading currently hard-limits the readable TimeseriesIndex range to INT32_MAX bytes and… · New
• This test uses POSIX open flags (O_WRONLY | O_CREAT | O_TRUNC) and fixed-width integer types… · New
• The per-test file name uses only TestInfo::name(), which can collide across different test suites… · New
• This test uses POSIX open flags (O_WRONLY | O_CREAT | O_TRUNC) and fixed-width integer types… · New
• Since this code path targets wide devices and can insert thousands of entries into name_to_index,… · New

5 resolved since last review

• Asserting chunks->size() == 1 can make this test brittle if chunking behavior changes (e.g.,…
• This test allocates MeasurementSchema with new and passes raw pointers, but it's not clear from…
• Allocating the entire read buffer (ti_buf) from PageArena means the raw bytes for the whole…
• The comment is inaccurate: GoogleTest typically runs all tests in a single process by default.…
• The comment says it deserializes \"every TimeseriesIndex\" in the range, but the implementation…

_{Review effort: Lite}