Ceph » rgw

Hi @Alexander Kazansky

Could you please elaborate how the `parquet` files with `object has a slash in name` are created for reproduction purposes?

like:

oid=2804fdd4-5ce0-4803-9588-5b622abc0ae2.10210173.6__shadow_process_launch/data/20230428_061954_01529_bzrj7-037aaaa7-8448-43a5-a4a4-49821d048de4.parquet.2~fvn7iKYGBa20LXxLl1AjL8RIILTl-LT.2_3

in a reproduction attempt of an object with slashes like `s3://bkt/xxx/8M.dat` the cached file name (`oid`) does not contain slashes:

❯ s3cmd ls --human --recursive s3://bkt
2023-05-22 15:24     8M  s3://bkt/8M.dat
2023-05-22 15:26     8M  s3://bkt/xxx/8M.dat
❯ ls -ltrh /mnt/nvme/rgw_datacache
total 8.0M
-rw-r--r-- 1 root root 4.0M May 22 15:25 1d67f18b-f606-4327-822e-79a16373869d.4204.2__shadow_.Il5abVEb3Tu3XKIGSdZ5-eLQj90hrU8_1
-rw-r--r-- 1 root root 4.0M May 22 15:27 1d67f18b-f606-4327-822e-79a16373869d.4204.2__shadow_.PPXgs45aR6ixGWzMIxcvRwy5IMa2vRl_1

Mark Kogan wrote:

Hi @Alexander Kazansky

Hi Mark,

Could you please elaborate how the `parquet` files with `object has a slash in name` are created for reproduction purposes?

like:
[...]

in current time we use next structure for store a our iceberg tables
bucket for store some table union by logical principles
2804fdd4-5ce0-4803-9588-5b622abc0ae2.10210173
prefix (folder) process_launch is name of the table
and "/data" is prefix (folder) for sore data files composing this table. by default iceberg table stores data files in location/data/ path and metadata files in location/metadata/

in a reproduction attempt of an object with slashes like `s3://bkt/xxx/8M.dat` the cached file name (`oid`) does not contain slashes:

[...]

i find what my problem true for parquet files. If i upload a random generated file or a tar.gz archive, rgw hashed name before write the file in cache path, also as in your example.
for reproduction problem try get any parquete file, for example from https://www.nyc.gov/site/tlc/about/tlc-trip-record-data.page

in current time we use next structure for store a our iceberg tables
bucket for store some table union by logical principles
2804fdd4-5ce0-4803-9588-5b622abc0ae2.10210173
prefix (folder) process_launch is name of the table
and "/data" is prefix (folder) for sore data files composing this table. by default iceberg table stores data files in location/data/ path and metadata files in location/metadata/

Thank you for detailing,
Is it possible to please ask for s3cmd ls --recursive on a such bucket ?
and a rgw log (debug_rgw = 20) file (like above) but of a whole GET operation please so can observe the complete GET op HTTP headers

Thanks

Thank you very much for the logs, also able to repro something that looks similar with Trino
debugging ...