Thanks Daniel for your thorough explanation. I understand your concern now and agree with this point.
I feel like we're adding additional information that isn't accurate and > doesn't add a whole lot of value. > Actually, maintenance tasks scheduling doesn't require an accurate timestamp, which also means I can use the timestamp when the service receives the commit report. Cancel this vote and close the PR. Thanks Eduard and Yufei for reviewing the PR. Manu On Wed, May 21, 2025 at 2:39 AM Daniel Weeks <dwe...@apache.org> wrote: > Hey Manu, > > I didn't see a discuss thread on this topic, so I'll add my concerns here. > > The issue I have is around the fidelity of what we're using as a commit > timestamp. I feel like we're adding additional information that isn't > accurate and doesn't add a whole lot of value. > > The metrics already include the snapshot id, so it's possible to lookup > the snapshot timestamp in the metadata, but that timestamp is also not a > canonical commit timestamp. Iceberg doesn't rely on or require that the > timestamp accurately reflect the point at which the commit was effected (in > fact you can have reads for older snapshots happen before the next snapshot > is live for other readers). This means the value cannot be relied on for > accurate sequencing of changes. This being the case, time of receipt for > the request is roughly equivalent. > > Given this, I would currently vote -0 (with a strong preference for not > including additional information where it's not necessary). > > -Dan > > > > On Fri, May 16, 2025 at 8:11 AM Manu Zhang <owenzhang1...@gmail.com> > wrote: > >> Thanks Yufei. I'm still seeking more votes here. >> >> Manu >> >> On Wed, May 14, 2025 at 1:37 AM Yufei Gu <flyrain...@gmail.com> wrote: >> >>> +1 I'm OK to add it as long as it's optional. >>> >>> Yufei >>> >>> >>> On Mon, May 12, 2025 at 8:47 PM Manu Zhang <owenzhang1...@gmail.com> >>> wrote: >>> >>>> Hi all, >>>> >>>> The background is that we schedule maintenance jobs based on commit >>>> reports for Iceberg tables, and we want to know *when commits happen*. >>>> Adding timestamp to the commit report would save us from loading >>>> metadata of every table from the filesystem. >>>> >>>> Please take a look at the PR and cast your vote. Feel free to ping me >>>> if you have any questions. >>>> >>>> Thanks, >>>> Manu >>>> >>>> >>>> On Fri, May 9, 2025 at 12:01 AM Manu Zhang <owenzhang1...@gmail.com> >>>> wrote: >>>> >>>>> Hi all, >>>>> >>>>> I'd like to start a vote to add commit timestamp `timestamp-millis` to >>>>> CommitReport in PR 12990 >>>>> <https://github.com/apache/iceberg/pull/12990/>. >>>>> The timestamp info is valuable to schedule maintenance jobs, but >>>>> currently we need to look it up from the snapshot metadata table. >>>>> >>>>> Please take a look at the PR and vote in the next 72 hours. >>>>> >>>>> [ ] +1 ok to add the field >>>>> [ ] +0 >>>>> [ ] -1 I have concerns or there's a better alternative >>>>> >>>>> Thanks, >>>>> Manu >>>>> >>>>
