horaedb icon indicating copy to clipboard operation
horaedb copied to clipboard

add inverted index of timeseries

Open chunshao90 opened this issue 3 years ago • 3 comments

Description

Add the inverted index of the timeseries to optimize the time series query for a small number of timelines. Similar to tsi.

Proposal

Build: Option 1: Add index to memtable and build it when data is written. Option 2: Build when memtable flushes to sst.

Storage: Option 1: Store in parquet metadata. Option 2: Store in separate file.

chunshao90 avatar Jun 16 '22 06:06 chunshao90

iox 已经去掉了 inverted index, 全靠 datafusion 执行引擎 table scan

aierui avatar Jun 18 '22 09:06 aierui

iox 已经去掉了 inverted index, 全靠 datafusion 执行引擎 table scan

In some scenes(like point query used in alerts), batch scan may be much more expensive compared to scaning/getting filterd rows with the inverted index. And now inverted index feature is currently in the poc stage, and we will decide the next step based on specific performance tests.

chunshao90 avatar Jun 20 '22 03:06 chunshao90

@chunshao90 Thank your reply, I'm looking forward to this feature.

aierui avatar Jun 20 '22 10:06 aierui

Duplicated with https://github.com/CeresDB/ceresdb/issues/628

jiacai2050 avatar Mar 02 '23 13:03 jiacai2050