Optimize timeseries metadata read and fix table size accounting

[shuwenwei]

This PR includes two related fixes around TsFile metadata handling.

First, it adds offset-based readTimeseriesMetadata overloads in TsFileSequenceReader. When callers already have the current device's measurement metadata node offset from TsFileDeviceIterator, the reader can pass the long[] offset directly and avoid searching the device metadata index again. Existing overloads continue to pass null and keep the original lookup behavior.

Second, it fixes table size accounting in TsFileIOWriter. The previous logic did not initialize the first table name before handling table switches, so metadata size for the first table could be missed, especially when the first table only had one device.

Changes

• Add readTimeseriesMetadata overloads that accept a nullable long[] deviceMetadataIndexNodeOffset.
• Use the provided device measurement metadata node offset as a fast path when it is not null.
• Keep existing overloads unchanged by forwarding them with null.
• Synchronize large metadata reads that move the shared tsFileInput position.
• Initialize the first table name during metadata index construction in TsFileIOWriter.
• Add a TsFileDeviceIteratorTest case that reads timeseries metadata with and without the iterator-provided offset.
• Add calculateTableSize2 to cover table size accounting when the first table has only one device.

Validation

Not run locally.

[Copilot]

Pull request overview

This PR improves TsFile metadata handling by adding an offset-based fast path for reading timeseries metadata (avoiding redundant device-index lookups when the caller already has the measurement-node range), and fixes table size accounting for the first table during metadata index construction.

Changes:

• Added readTimeseriesMetadata overloads in TsFileSequenceReader that accept an optional long[] measurement-node offset range to skip device-index searching.
• Synchronized large metadata reads that use tsFileInput.position(...) and fixed first-table initialization for table size accounting in TsFileIOWriter.
• Added tests covering the iterator-provided offset path and the “first table has only one device” table size accounting case.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 5 comments.

File	Description
java/tsfile/src/main/java/org/apache/tsfile/read/TsFileSequenceReader.java	Adds offset-based metadata read overloads and synchronizes some large metadata reads.
java/tsfile/src/main/java/org/apache/tsfile/write/writer/TsFileIOWriter.java	Initializes first table name early so first-table metadata size is accounted during table switches.
java/tsfile/src/test/java/org/apache/tsfile/read/TsFileDeviceIteratorTest.java	Adds a test reading timeseries metadata with/without iterator-provided measurement-node offsets.
java/tsfile/src/test/java/org/apache/tsfile/write/TsFileWriteApiTest.java	Adds a table size accounting test for a first table with a single device.

Comments suppressed due to low confidence (1)

java/tsfile/src/main/java/org/apache/tsfile/read/TsFileSequenceReader.java:1701

• Only the single-boolean overload of generateMetadataIndexUsingTsFileInput is synchronized, but the 2-boolean overload (which directly moves tsFileInput.position) is still callable and is invoked directly elsewhere in this class (e.g., from TimeseriesMetadataIterator). To actually make large-metadata reads thread-safe, the underlying 2-boolean method should also be synchronized (or all call sites should go through a synchronized wrapper).

  private synchronized void generateMetadataIndexUsingTsFileInput(
      IMetadataIndexEntry metadataIndex,
      long start,
      long end,
      IDeviceID deviceId,

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.