Incorrect timestamps returns by MultiGet when hit row cache

[burtonli]

Note: Please use Issues only for bug reports. For questions, discussions, feature requests, etc. post to dev group: https://groups.google.com/forum/#!forum/rocksdb or https://www.facebook.com/groups/rocksdb.dev

Expected behavior

Timestamps returned by MultiGet should be matched with timestamps of WriteOptions when records were written .

Actual behavior

When hit record cache, the returned timestamps are matched with the timestamp of ReadOptions.

Steps to reproduce the behavior

Initial DB with enough capacity for row_cache.

1. Write record A with V1+ timestamp "1000"
2. Write record A with V2 + timestamp "2000"
3. Read record A with timestamp "1500", it returns V1 + timestamp with "1000" correctly.
4. Read record A with timestamp "1500" again (hit row cache), it returns V1 + timestamp "1500" ("1000" is expected)

Root cause:
In the implementation of TableCache::Get(), when the user query key hit row cache, it call replayGetContextLog() by directly passing user query key, where it will extract query timestamp (1500) instead of timestamp for the record (1000).
@riversand963

Internal ref: T86715827

[jay-zhuang]

Yeah, currently timestamp + row_cache is not supported yet.

[riversand963]

Hi @burtonli , would you like to contribute this feature?

[burtonli]

I don't have plan in short future.

[thejchap]

i am happy to take a look at this one

[burtonli]

Thank you @thejchap , do we have any finding on this?

[thejchap]

@burtonli thanks for following up. got derailed with team selection and onboarding. fully onboarded now and easing back in to normal day to day life. i plan to revisit this next week.

[sherriiiliu]

Hi @thejchap , is there any update on this bug fix?

[sherriiiliu]

Hi @thejchap , is there any update on this bug fix?

Kindly ping again @thejchap , is the bug fixed?

[cz2h]

Hi @thejchap , any updates on this issue? If this issue is still up for grad, can I work on this?

[thejchap]

@cz2h all yours