Add chdb solution #131
Add chdb solution #131
Conversation
The cbdb implementation is based on polars python code. Current approach is using connection + cursor.
Draft for joining code is finished
Grouping queries were prepared and also added flush=True for all printing operations with LIMIT 3
Now groupings code is working
fix nc field
Initial working version of chdb for join
Proper logic for on_disk and na_flag identification was added.
fix for local run
Added settings and fix mergeTree for session branch
added latest query and max_threads
Added settings for threads
on M1 Pro chip - 8 threads showing better perf
| conn.query("DROP TABLE IF EXISTS ans") | ||
| gc.collect() | ||
| if compress: | ||
| time.sleep(60) |
There was a problem hiding this comment.
why are we putting a sleep between the two queries?
There was a problem hiding this comment.
The processing logic of chDB is similar to that of ClickHouse, so here, just like with ClickHouse, the same handling has been applied for the 50G large dataset to avoid OOM.
I added the following comments in the test script:
# It will take some time for memory freed by Memory engine to be returned back to the system.
# Without a sleep we might get a MEMORY_LIMIT exception during the second run of the query.
# It is done only when compress is true because this variable is set to true only for the largest dataset.
| QUERY=f"CREATE TABLE ans ENGINE = {query_engine} AS SELECT id1, sum(v1) AS v1 FROM db_benchmark.x GROUP BY id1 {settings}" | ||
| conn.query(QUERY) | ||
| nr = int(str(conn.query("SELECT count(*) AS cnt FROM ans")).strip()) | ||
| nc = len(str(conn.query("SELECT * FROM ans LIMIT 0", "CSVWITHNAMES")).split(',')) |
There was a problem hiding this comment.
What is the significant of "CSVWITHnAMES" here? Just curious
There was a problem hiding this comment.
Oh, is it just a way to get the number of columns?
There was a problem hiding this comment.
Yes, it's just to be able to get the number of columns.
chdb/monitor_ram.py
Outdated
| import time | ||
| import sys | ||
|
|
||
| solution = str(sys.argv[1]) |
There was a problem hiding this comment.
does this file need to be included? Maybe it could be in utils with the idea all solutions can use it?
There was a problem hiding this comment.
This file should be unused and identical to monitor_ram.py in the polars directory, so I have deleted it.
There was a problem hiding this comment.
Yes, as I ported the solution from the polars folder - this file was taken from there.
Thank you very much. I have reverted the changes made to time.csv and logs.csv. |
|
Nice timing for new version of the benchmark to compare with newest DuckDB 1.4.0 |
|
@wudidapaopao seems like the regression test is failing for chdb? Can you take a look? The conflict is just that I removed pydatatable from the regression runner. The failing Julia solutions is known. Eventually I will fix those as well |
|
@Tmonster I've resolved the code conflicts and attempted to fix the failing regression tests, with chk output added. Since I couldn't reproduce the issue locally, would it be convenient to re-trigger the regression testing? I'm hoping this fix will work. |
|
Awesome thanks! |
3 participants
This PR adds the chdb solution.
I've updated the test results for the
0.5Gand5Gdatasets, which were run on my local laptop, intime.csvandlogs.csv. These results were not generated on a standard machine. They are only intended to demonstrate the final output of the test script.Please feel free to contact me if there are any issues or missing information.