ycycse opened a new pull request, #769:
URL: https://github.com/apache/tsfile/pull/769
Main changes:
- build dataset metadata from `get_timeseries_metadata()` instead of
scanning full table data during
`TsFileSeriesReader` initialization
- use table-model `query_table_by_row(...)` to support lazy `Timeseries`
position access (`series[i]`, small slices)
- optimize overlapping-shard position reads with chunked k-way merge,
avoiding full timestamp materialization for
small reads
- remove the incorrect device-level timestamp-axis assumption and keep
duplicate timestamp validation on logical-
series paths
- fix `query_table_by_row(..., tag_filter=...)` tag-filter handle
ownership in Python bindings
## C++ note
This PR also keeps a small C++ reader fix in `tsfile_reader.cc/.h`.
After `query_table_by_row(...)` was introduced, the same `TsFileReader`
could mix row-mode and batch-mode table
queries. Reusing one `TableQueryExecutor` across those mode switches could
leave the reader in the wrong mode and
break later batch reads.
To avoid that, `TsFileReader` now recreates `TableQueryExecutor` when the
requested table-query mode/configuration
changes.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
