feat: try to improve object storage io read

[BohuTANG]

I hereby agree to the terms of the CLA available at: https://databend.rs/dev/policies/cla/

Summary

Try to merge neighboring io operations using RangeMerger

• If the distance between two IO ranges to be read in one file is less than storage_io_min_bytes_for_seek, then Databend sequentially reads a range of files that contains both ranges, thus avoiding extra seek.
• This only works for the remote read.

Introduce two remote read IO settings

• storage_io_min_bytes_for_seek: If the distance between two IO requests to be read in one file is less than storage_io_min_bytes_for_seek, merge them into one, the default value is 512Bytes.
• storage_io_max_page_bytes_for_read: The maximum bytes of one IO request to read. Default the value is 512KB

Introduce remote IO request perf counter

• remote_io_seeks: the raw range numbers before IO merged
• remote_io_seeks_after_merged: the ranges need to read after IO merged
• remote_io_read_bytes: the bytes read before IO merged
• remote_io_read_bytes_after_merged: the bytes need to read after IO merged
• remote_io_read_milliseconds: read cost in milliseconds

Case1:

SELECT * FROM hits WHERE URL LIKE '%google%' ORDER BY EventTime LIMIT 10 ignore_result;

-- No merged IO (set storage_io_min_bytes_for_seek=0;)
mysql> select metric, value  from system.metrics where metric like '%remote_io%' order by metric;
+----------------------------------------+---------------+
| metric                                 | value         |
+----------------------------------------+---------------+
| fuse_remote_io_read_bytes              | 15205646411.0 |
| fuse_remote_io_read_bytes_after_merged | 15205646411.0 |
| fuse_remote_io_read_milliseconds       | 546271.0      |
| fuse_remote_io_read_parts              | 1490.0        |
| fuse_remote_io_seeks                   | 77607.0       |
| fuse_remote_io_seeks_after_merged      | 77607.0       |
+----------------------------------------+---------------+

-- Merged IO (set storage_io_min_bytes_for_seek=48)
mysql> select metric, value  from system.metrics where metric like '%remote_io%' order by metric;
+----------------------------------------+---------------+
| metric                                 | value         |
+----------------------------------------+---------------+
| fuse_remote_io_read_bytes              | 15205646411.0 |
| fuse_remote_io_read_bytes_after_merged | 15208201850.0 |
| fuse_remote_io_read_milliseconds       | 446209.0      |
| fuse_remote_io_read_parts              | 1490.0        |
| fuse_remote_io_seeks                   | 77607.0       |
| fuse_remote_io_seeks_after_merged      | 13904.0       |
+----------------------------------------+---------------+

For this query, the total IO time was from 546271.0ms to 446209.0ms(after Merged IO).

Case2:

mysql> select * from hits ignore_result;
Empty set (19.74 sec)
Read 99997497 rows, 75.78 GiB in 19.733 sec., 5.07 million rows/sec., 3.84 GiB/sec.

mysql> select metric, value  from system.metrics where metric like '%remote_io%' order by metric;
+----------------------------------------+---------------+
| metric                                 | value         |
+----------------------------------------+---------------+
| fuse_remote_io_read_bytes              | 15412179252.0 |
| fuse_remote_io_read_bytes_after_merged | 15412179252.0 |
| fuse_remote_io_read_milliseconds       | 375970.0      |
| fuse_remote_io_read_parts              | 751.0         |
| fuse_remote_io_seeks                   | 78855.0       |
| fuse_remote_io_seeks_after_merged      | 78855.0       |
+----------------------------------------+---------------+


mysql> select * from hits ignore_result;
Empty set (19.71 sec)
Read 99997497 rows, 75.78 GiB in 19.701 sec., 5.08 million rows/sec., 3.85 GiB/sec.

mysql> select metric, value  from system.metrics where metric like '%remote_io%' order by metric;
+----------------------------------------+---------------+
| metric                                 | value         |
+----------------------------------------+---------------+
| fuse_remote_io_read_bytes              | 15412179252.0 |
| fuse_remote_io_read_bytes_after_merged | 15414801174.0 |
| fuse_remote_io_read_milliseconds       | 337415.0      |
| fuse_remote_io_read_parts              | 751.0         |
| fuse_remote_io_seeks                   | 78855.0       |
| fuse_remote_io_seeks_after_merged      | 13461.0       |
+----------------------------------------+---------------+

Bench

Nightly-162 vs. This PR

Closes #9308

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

1 Ignored Deployment

Name	Status	Preview	Updated
databend	⬜️ Ignored (Inspect)		Dec 26, 2022 at 2:48AM (UTC)

[BohuTANG]

@mergify update

[mergify]

update

✅ Branch has been successfully updated

[BohuTANG]

@mergify update

[mergify]

update

✅ Branch has been successfully updated

[BohuTANG]

@mergify update

[mergify]

update

✅ Branch has been successfully updated