matanper commented on PR #69278:
URL: https://github.com/apache/doris/pull/69278#issuecomment-6288264924
@gavinchou thanks for taking a look!
this PR doesn't introduce the layout. path_version = 1
(data/{shard}/{tablet_id}/{rowset_id}/{seg}.dat, shard =
murmur_hash64A(tablet_id) % shard_num) has existed since #35953
It's implemented in BE StorageResource, exposed as the
path_version/shard_num properties of CREATE STORAGE VAULT, and works for HDFS
vaults.
As for why we want v1 on S3: S3 scales request throughput per key prefix (on
the order of 3,500 PUT/s per partitioned prefix) and splits partitions by key
range over time. Tablet ids are allocated sequentially, so after a TRUNCATE, a
new table or a new partition, all new writes go to data/<adjacent tablet
ids>/..., one narrow key range, until S3 re-partitions. Under heavy ingest
(about 23 Gbit/s of uploads per BE) we got tens of thousands of SlowDown/503 on
UploadPart, and loads failed. A hash-based first component spreads new tablets
across shard_num prefixes from the first write.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
