precompute stats for merge multivalue index + disable Line encoding for
multivalue index. That combination allows to skip the first estimation
pass. This gives up to 2x on merge performance on multivalue indices.
This change may decrease compression as Line is
very good compressible for documents, which have a fixed amount of
values in each doc. The line codec should be replaced.
```
merge_multi_and_multi Avg: 22.7880ms (-47.15%) Median: 22.5469ms (-47.38%) [22.3691ms .. 25.8392ms]
merge_dense_and_dense Avg: 14.4398ms (+2.18%) Median: 14.2465ms (+0.74%) [14.1620ms .. 16.1270ms]
merge_sparse_and_sparse Avg: 10.6559ms (+1.10%) Median: 10.6318ms (+0.91%) [10.5527ms .. 11.2848ms]
merge_sparse_and_dense Avg: 12.4886ms (+1.52%) Median: 12.4044ms (+0.84%) [12.3261ms .. 13.9439ms]
merge_multi_and_dense Avg: 25.6686ms (-45.56%) Median: 25.4851ms (-45.84%) [25.1618ms .. 27.6226ms]
merge_multi_and_sparse Avg: 24.3278ms (-47.00%) Median: 24.1917ms (-47.34%) [23.7159ms .. 27.0513ms]
```
* add method to fetch block of first vals in columnar
add method to fetch block of first vals in columnar (this is way faster
than single calls for full columns)
add benchmark
fix import warnings
```
test bench_get_block_first_on_full_column ... bench: 56 ns/iter (+/- 26)
test bench_get_block_first_on_full_column_single_calls ... bench: 311 ns/iter (+/- 6)
test bench_get_block_first_on_multi_column ... bench: 378 ns/iter (+/- 15)
test bench_get_block_first_on_multi_column_single_calls ... bench: 546 ns/iter (+/- 13)
test bench_get_block_first_on_optional_column ... bench: 291 ns/iter (+/- 6)
test bench_get_block_first_on_optional_column_single_calls ... bench: 362 ns/iter (+/- 8)
```
* use remainder
* handle ip adresses in term aggregation
Stores IpAdresses during the segment term aggregation via u64 representation
and convert to u128(IpV6Adress) via downcast when converting to intermediate results.
Enable Downcasting on `ColumnValues`
Expose u64 variant for u128 encoded data via `open_u64_lenient` method.
Remove lifetime in VecColumn, to avoid 'static lifetime requirement coming
from downcast trait.
* rename method
Spotted in `range_date_histogram` query in quickwit benchmark:
5% of time copying docs around, which is not needed in the full index case
remove Column to ColumnIndex deref
* Fix bug that can cause `get_docids_for_value_range` to panic.
When `selected_docid_range.end == num_rows`, we would get a panic
as we try to access a non-existing blockmeta.
This PR accepts calls to rank with any value.
For any value above num_rows we simply return non_null_rows.
Fixes#2293
* add tests, merge variables
---------
Co-authored-by: Pascal Seitz <pascal.seitz@gmail.com>
* read path for new fst based index
* implement BlockAddrStoreWriter
* extract slop/derivation computation
* use better linear approximator and allow negative correction to approximator
* document format and reorder some fields
* optimize single block sstable size
* plug backward compat
* add fields_metadata to SegmentReader, add columnar docs
* use schema to resolve field, add test
* normalize paths
* merge for FieldsMetadata, add fields_metadata on Index
* Update src/core/segment_reader.rs
Co-authored-by: Paul Masurel <paul@quickwit.io>
* merge code paths
* add Hash
* move function oustide
---------
Co-authored-by: Paul Masurel <paul@quickwit.io>
* add missing aggregation part 2
Add missing support for:
- Mixed types columns
- Key of type string on numerical fields
The special aggregation is slower than the integrated one in TermsAggregation and therefore not
chosen by default, although it can cover all use cases.
* simplify, add num_docs to empty
* lazy columnar merge
This is the first part of addressing #3633
Instead of loading all Column into memory for the merge, only the current column_name
group is loaded. This can be done since the sstable streams the columns lexicographically.
* refactor
* add rustdoc
* replace iterator with BTreeMap
* Do some Clippy- and Cargo-related boy-scouting.
* Add BytesFilterCollector to support filtering based on a bytes fast field
This is basically a copy of the existing FilterCollector but modified and
specialised to work on a bytes fast field.
* Changed semantics of filter collectors to consider multi-valued fields
* switch to ms in histogram for date type
switch to ms in histogram, by adding a normalization step that converts
to nanoseconds precision when creating the collector.
closes#2028
related to #2026
* add missing unit long variants
* use single thread to avoid handling test case
* fix docs
* revert CI
* cleanup
* improve docs
* Update src/aggregation/bucket/histogram/histogram.rs
Co-authored-by: Paul Masurel <paul@quickwit.io>
---------
Co-authored-by: Paul Masurel <paul@quickwit.io>
Applied this command to the code, making it a bit shorter and slightly
more readable.
```
cargo +nightly clippy --all-features --benches --tests --workspace --fix -- -A clippy::all -W clippy::uninlined_format_args
cargo +nightly fmt --all
```
* compress sstable with zstd
* add some details to sstable readme
* compress only block which benefit from it
* multiple changes to sstable
make compression optional
use OwnedBytes instead of impl Read in sstable, required for next point
use zstd bulk api, which is much faster on small records
* cleanup and use bulk api for compression
* use dedicated byte for compression
* switch block len and compression flag
* change default zstd level in sstable
* Added proptest on columnar merge with a shuffle
Made column serialization more explicit.
Bugfix when a bytes column is missing, and with a shuffle.
Improved the cardinality detection logic / column detection.
* Code review
* CR comments
* Following CR
* Faster range queries
This PR does several changes
- ip compact space now uses u32
- the bitunpacker now gets a get_batch function
- we push down range filtering, removing GCD / shift in the bitpacking
codec.
- we rely on AVX2 routine to do the filtering.
* Apply suggestions from code review
* Apply suggestions from code review
* CR comments
* document a new sstable format
* add support for changing target block size
* use new format for sstable index
* handle sstable version errror
* use very small blocks for proptests
* add a footer structure
