[WIP]Test orc prefetch

[taiyang-li]

Changelog category (leave one):

• New Feature
• Experimental Feature
• Improvement
• Performance Improvement
• Backward Incompatible Change
• Build/Testing/Packaging Improvement
• Documentation (changelog entry is not required)
• Critical Bug Fix (crash, LOGICAL_ERROR, data loss, RBAC)
• Bug Fix (user-visible misbehavior in an official stable release)
• CI Fix or Improvement (changelog entry is not required)
• Not for changelog (changelog entry is not required)

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

...

Documentation entry for user-facing changes

• Documentation is written (mandatory for new features)

Information about CI checks: https://clickhouse.com/docs/en/development/continuous-integration/

CI Settings (Only check the boxes if you know what you are doing):

• Allow: All Required Checks
• Allow: Stateless tests
• Allow: Stateful tests
• Allow: Integration Tests
• Allow: Performance tests
• Allow: All Builds
• Allow: batch 1, 2 for multi-batch jobs
• Allow: batch 3, 4, 5, 6 for multi-batch jobs

---

• Exclude: Style check
• Exclude: Fast test
• Exclude: All with ASAN
• Exclude: All with TSAN, MSAN, UBSAN, Coverage
• Exclude: All with aarch64, release, debug

---

• Run only fuzzers related jobs (libFuzzer fuzzers, AST fuzzers, etc.)
• Exclude: AST fuzzers

---

• Do not test
• Woolen Wolfdog
• Upload binaries for special builds
• Disable merge-commit
• Disable CI cache

[taiyang-li]

with prefetch

select rip, rtime, country, os, event from hdfs('hdfs://bigocluster/user/hive/warehouse/test.orc') settings max_threads = 1, max_parsing_threads = 1, remote_filesystem_read_prefetch = 1, remote_read_min_bytes_for_seek = 524288 format Null; 
0 rows in set. Elapsed: 8.950 sec. Processed 780.68 thousand rows, 716.07 MB (87.23 thousand rows/s., 80.01 MB/s.)
0 rows in set. Elapsed: 5.397 sec. Processed 780.68 thousand rows, 716.07 MB (144.64 thousand rows/s., 132.67 MB/s.)
0 rows in set. Elapsed: 7.356 sec. Processed 780.68 thousand rows, 716.07 MB (106.13 thousand rows/s., 97.34 MB/s.)
0 rows in set. Elapsed: 5.735 sec. Processed 780.68 thousand rows, 716.07 MB (136.12 thousand rows/s., 124.86 MB/s.)
0 rows in set. Elapsed: 5.401 sec. Processed 780.68 thousand rows, 716.07 MB (144.54 thousand rows/s., 132.58 MB/s.)
0 rows in set. Elapsed: 5.010 sec. Processed 780.68 thousand rows, 716.07 MB (155.82 thousand rows/s., 142.93 MB/s.)
0 rows in set. Elapsed: 4.955 sec. Processed 780.68 thousand rows, 716.07 MB (157.55 thousand rows/s., 144.52 MB/s.)
0 rows in set. Elapsed: 5.907 sec. Processed 780.68 thousand rows, 716.07 MB (132.16 thousand rows/s., 121.22 MB/s.)
0 rows in set. Elapsed: 5.417 sec. Processed 780.68 thousand rows, 716.07 MB (144.12 thousand rows/s., 132.19 MB/s.)
0 rows in set. Elapsed: 5.649 sec. Processed 780.68 thousand rows, 716.07 MB (138.21 thousand rows/s., 126.77 MB/s.)

后续：经过数轮优化：最快可达到4.2s，加速1.47x。优化包括

• 每次prefetch之前，淘汰过时的buffer，以减少page fault
• orc DataBuffer调用resize时会调用memset, 开销比较大，已去除
• 限制range合并的上限值，充分利用io并发。

without prefetch

select rip, rtime, country, os, event from hdfs('hdfs://bigocluster/user/hive/warehouse/test.orc') settings max_threads = 1, max_parsing_threads = 1, remote_filesystem_read_prefetch = 0, remote_read_min_bytes_for_seek = 524288 format Null; 
0 rows in set. Elapsed: 11.423 sec. Processed 780.68 thousand rows, 716.07 MB (68.34 thousand rows/s., 62.69 MB/s.)
0 rows in set. Elapsed: 13.590 sec. Processed 780.68 thousand rows, 716.07 MB (57.44 thousand rows/s., 52.69 MB/s.)
0 rows in set. Elapsed: 7.953 sec. Processed 780.68 thousand rows, 716.07 MB (98.16 thousand rows/s., 90.04 MB/s.)
0 rows in set. Elapsed: 6.614 sec. Processed 780.68 thousand rows, 716.07 MB (118.04 thousand rows/s., 108.27 MB/s.)
0 rows in set. Elapsed: 7.343 sec. Processed 780.68 thousand rows, 716.07 MB (106.32 thousand rows/s., 97.52 MB/s.)
0 rows in set. Elapsed: 6.905 sec. Processed 780.68 thousand rows, 716.07 MB (113.06 thousand rows/s., 103.70 MB/s.)
0 rows in set. Elapsed: 6.442 sec. Processed 780.68 thousand rows, 716.07 MB (121.18 thousand rows/s., 111.15 MB/s.)
0 rows in set. Elapsed: 6.178 sec. Processed 780.68 thousand rows, 716.07 MB (126.36 thousand rows/s., 115.90 MB/s.)
0 rows in set. Elapsed: 6.471 sec. Processed 780.68 thousand rows, 716.07 MB (120.64 thousand rows/s., 110.66 MB/s.)
0 rows in set. Elapsed: 7.440 sec. Processed 780.68 thousand rows, 716.07 MB (104.93 thousand rows/s., 96.25 MB/s.)

[taiyang-li]

realtime flamegraph

with prefetch

without prefetch

[taiyang-li]

Close becuase of ClickHouse#70534