I have built a fuzz test and it crashes heavily in seconds and will OOM shortly after.
It seems like supporting Parquet is basically a completely open way to crash the
server if you can upload a file and run s3 select on it.
Until Parquet is more hardened it is DISABLED by default since hostile
crafted input can easily crash the server.
If you are in a controlled environment where it is safe to assume no hostile
content can be uploaded to your cluster you can safely enable Parquet.
To enable Parquet set the environment variable `MINIO_API_SELECT_PARQUET=on`
while starting the MinIO server.
Furthermore, we guard parquet by recover functions.
enable linter using golangci-lint across
codebase to run a bunch of linters together,
we shall enable new linters as we fix more
things the codebase.
This PR fixes the first stage of this
cleanup.
The SQL parser as it stands right now ignores alias for aggregate
result, e.g. `SELECT COUNT(*) AS thing FROM s3object` doesn't actually
return record like `{"thing": 42}`, it returns a record like `{"_1": 42}`.
Column alias for aggregate result is supported in AWS's S3 Select, so
this commit fixes that by respecting the `expr.As` in the expression.
Also improve test for S3 select
On top of testing a simple `SELECT` query, we want to test a few more
"advanced" queries (e.g. aggregation).
Convert existing tests into table driven tests[1], and add the new test
cases with "advanced" queries into them.
[1] - https://github.com/golang/go/wiki/TableDrivenTests
Batching records into a single SQL Select message in the response
leads to significant speed up as the message header overhead is made
negligible.
This change leads to a speed up of 3-5x for queries that select many
small records.
This improves the performance of certain queries dramatically,
such as 'count(*)' etc.
Without this PR
```
~ time mc select --query "select count(*) from S3Object" myminio/sjm-airlines/star2000.csv.gz
2173762
real 0m42.464s
user 0m0.071s
sys 0m0.010s
```
With this PR
```
~ time mc select --query "select count(*) from S3Object" myminio/sjm-airlines/star2000.csv.gz
2173762
real 0m17.603s
user 0m0.093s
sys 0m0.008s
```
Almost a 250% improvement in performance. This PR avoids a lot of type
conversions and instead relies on raw sequences of data and interprets
them lazily.
```
benchcmp old new
benchmark old ns/op new ns/op delta
BenchmarkSQLAggregate_100K-4 551213 259782 -52.87%
BenchmarkSQLAggregate_1M-4 6981901985 2432413729 -65.16%
BenchmarkSQLAggregate_2M-4 13511978488 4536903552 -66.42%
BenchmarkSQLAggregate_10M-4 68427084908 23266283336 -66.00%
benchmark old allocs new allocs delta
BenchmarkSQLAggregate_100K-4 2366 485 -79.50%
BenchmarkSQLAggregate_1M-4 47455492 21462860 -54.77%
BenchmarkSQLAggregate_2M-4 95163637 43110771 -54.70%
BenchmarkSQLAggregate_10M-4 476959550 216906510 -54.52%
benchmark old bytes new bytes delta
BenchmarkSQLAggregate_100K-4 1233079 1086024 -11.93%
BenchmarkSQLAggregate_1M-4 2607984120 557038536 -78.64%
BenchmarkSQLAggregate_2M-4 5254103616 1128149168 -78.53%
BenchmarkSQLAggregate_10M-4 26443524872 5722715992 -78.36%
```
Klaus Post
Anis Elleuch
Harshavardhana
Aditya Manthramurthy
Ryan Tam
Joe Stevens
kannappanr
Bala FA
Ashish Kumar Sinha
Praveen raj Mani
Raphael Randschau
Arjun Mishra