You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@polaris.apache.org by vignesh a <im...@gmail.com> on 2026/07/16 14:44:20 UTC
[VOTE] Return 503 Service Unavailable for TARGET_ENTITY_CONCURRENTLY_MODIFIED during table/view rename
Hi all,
Following the recent [DISCUSS] thread [1], I'd like to call a vote on the
HTTP status code to return when a table or view rename operation fails with
TARGET_ENTITY_CONCURRENTLY_MODIFIED.
*Background* 409 Conflict and 429 Too Many Requests were discussed as
alternatives. However:
- 409 conflicts with the Iceberg REST spec's existing meaning ("target
identifier already exists"), so clients would misinterpret it.
- 429 implies rate limiting and carries Retry-After semantics that don't
apply here.
The discussion converged on *503 Service Unavailable* as the cleanest
option - it is retryable and avoids the semantic conflicts above.
*Proposal * Return HTTP *503 Service Unavailable* when a table/view rename
fails with TARGET_ENTITY_CONCURRENTLY_MODIFIED. This signals a temporary
condition that clients can retry. Server-side automatic retries are *out of
scope* for this change and can be considered separately later.
My vote: *+1*
Please cast your vote by replying to this thread:
[ ] *+1 / Yes* -- Approve returning 503 [ ] *-1 / No* -- Reject this
proposal and resume discussion [ ] *0* -- Abstain
This vote will remain open for 72 hours and close at *14:00 UTC on Sunday,
19 July 2026*.
Thanks, *Vignesh*
[1] https://lists.apache.org/thread/tr8zh8121t2jb41s0q2yd9s73y2tp2tq
This change is implemented in PR #4646
<https://github.com/apache/polaris/pull/4646> .
Re: [VOTE] Return 503 Service Unavailable for TARGET_ENTITY_CONCURRENTLY_MODIFIED during table/view rename
Posted by Alexandre Dutra <ad...@apache.org>.
+1 for 503 + Retry-After header.
On Fri, Jul 17, 2026 at 11:27 AM Robert Stupp <sn...@snazy.de> wrote:
>
> +1 to returning 503 for this case, **with** the **Retry-After header** as
> discussed, so Iceberg REST clients have the expected retry signal for a
> non-idempotent request.
>
> Small process note: for future API-contract votes, it would be helpful to
> avoid running most of the vote window over a weekend when possible. This
> one seems narrow and already discussed, so I do not think it needs to
> restart, but weekday timing gives more people a practical chance to notice
> and respond.
>
> On Fri, Jul 17, 2026 at 10:42 AM Nándor Kollár <ko...@gmail.com>
> wrote:
>
> > +1 (non-binding)
> >
> > I think that '503 Service Unavailable' with a 'Retry-After' header
> > seems like the least bad option to me.
> >
> > Thanks,
> > Nandor
> >
> > vignesh a <im...@gmail.com> ezt írta (időpont: 2026. júl. 16.,
> > Cs, 16:44):
> > >
> > > Hi all,
> > >
> > > Following the recent [DISCUSS] thread [1], I'd like to call a vote on the
> > > HTTP status code to return when a table or view rename operation fails
> > with
> > > TARGET_ENTITY_CONCURRENTLY_MODIFIED.
> > >
> > > *Background* 409 Conflict and 429 Too Many Requests were discussed as
> > > alternatives. However:
> > >
> > > - 409 conflicts with the Iceberg REST spec's existing meaning ("target
> > > identifier already exists"), so clients would misinterpret it.
> > > - 429 implies rate limiting and carries Retry-After semantics that
> > don't
> > > apply here.
> > >
> > > The discussion converged on *503 Service Unavailable* as the cleanest
> > > option - it is retryable and avoids the semantic conflicts above.
> > >
> > > *Proposal * Return HTTP *503 Service Unavailable* when a table/view
> > rename
> > > fails with TARGET_ENTITY_CONCURRENTLY_MODIFIED. This signals a temporary
> > > condition that clients can retry. Server-side automatic retries are *out
> > of
> > > scope* for this change and can be considered separately later.
> > >
> > > My vote: *+1*
> > >
> > > Please cast your vote by replying to this thread:
> > >
> > > [ ] *+1 / Yes* -- Approve returning 503 [ ] *-1 / No* -- Reject this
> > > proposal and resume discussion [ ] *0* -- Abstain
> > >
> > > This vote will remain open for 72 hours and close at *14:00 UTC on
> > Sunday,
> > > 19 July 2026*.
> > >
> > > Thanks, *Vignesh*
> > >
> > > [1] https://lists.apache.org/thread/tr8zh8121t2jb41s0q2yd9s73y2tp2tq
> > >
> > > This change is implemented in PR #4646
> > > <https://github.com/apache/polaris/pull/4646> .
> >
> >
> >
> > --
> > Kollár Nándor
> >
Re: [VOTE] Return 503 Service Unavailable for TARGET_ENTITY_CONCURRENTLY_MODIFIED during table/view rename
Posted by Dmitri Bourlatchkov <di...@apache.org>.
Good point from Robert about the weekend.
Let's just extend the grace period until Tuesday next week maybe?
Cheers,
Dmitri.
On Fri, Jul 17, 2026 at 5:27 AM Robert Stupp <sn...@snazy.de> wrote:
> +1 to returning 503 for this case, **with** the **Retry-After header** as
> discussed, so Iceberg REST clients have the expected retry signal for a
> non-idempotent request.
>
> Small process note: for future API-contract votes, it would be helpful to
> avoid running most of the vote window over a weekend when possible. This
> one seems narrow and already discussed, so I do not think it needs to
> restart, but weekday timing gives more people a practical chance to notice
> and respond.
>
> On Fri, Jul 17, 2026 at 10:42 AM Nándor Kollár <ko...@gmail.com>
> wrote:
>
> > +1 (non-binding)
> >
> > I think that '503 Service Unavailable' with a 'Retry-After' header
> > seems like the least bad option to me.
> >
> > Thanks,
> > Nandor
> >
> > vignesh a <im...@gmail.com> ezt írta (időpont: 2026. júl. 16.,
> > Cs, 16:44):
> > >
> > > Hi all,
> > >
> > > Following the recent [DISCUSS] thread [1], I'd like to call a vote on
> the
> > > HTTP status code to return when a table or view rename operation fails
> > with
> > > TARGET_ENTITY_CONCURRENTLY_MODIFIED.
> > >
> > > *Background* 409 Conflict and 429 Too Many Requests were discussed as
> > > alternatives. However:
> > >
> > > - 409 conflicts with the Iceberg REST spec's existing meaning
> ("target
> > > identifier already exists"), so clients would misinterpret it.
> > > - 429 implies rate limiting and carries Retry-After semantics that
> > don't
> > > apply here.
> > >
> > > The discussion converged on *503 Service Unavailable* as the cleanest
> > > option - it is retryable and avoids the semantic conflicts above.
> > >
> > > *Proposal * Return HTTP *503 Service Unavailable* when a table/view
> > rename
> > > fails with TARGET_ENTITY_CONCURRENTLY_MODIFIED. This signals a
> temporary
> > > condition that clients can retry. Server-side automatic retries are
> *out
> > of
> > > scope* for this change and can be considered separately later.
> > >
> > > My vote: *+1*
> > >
> > > Please cast your vote by replying to this thread:
> > >
> > > [ ] *+1 / Yes* -- Approve returning 503 [ ] *-1 / No* -- Reject this
> > > proposal and resume discussion [ ] *0* -- Abstain
> > >
> > > This vote will remain open for 72 hours and close at *14:00 UTC on
> > Sunday,
> > > 19 July 2026*.
> > >
> > > Thanks, *Vignesh*
> > >
> > > [1] https://lists.apache.org/thread/tr8zh8121t2jb41s0q2yd9s73y2tp2tq
> > >
> > > This change is implemented in PR #4646
> > > <https://github.com/apache/polaris/pull/4646> .
> >
> >
> >
> > --
> > Kollár Nándor
> >
>
Re: [VOTE] Return 503 Service Unavailable for TARGET_ENTITY_CONCURRENTLY_MODIFIED during table/view rename
Posted by Robert Stupp <sn...@snazy.de>.
+1 to returning 503 for this case, **with** the **Retry-After header** as
discussed, so Iceberg REST clients have the expected retry signal for a
non-idempotent request.
Small process note: for future API-contract votes, it would be helpful to
avoid running most of the vote window over a weekend when possible. This
one seems narrow and already discussed, so I do not think it needs to
restart, but weekday timing gives more people a practical chance to notice
and respond.
On Fri, Jul 17, 2026 at 10:42 AM Nándor Kollár <ko...@gmail.com>
wrote:
> +1 (non-binding)
>
> I think that '503 Service Unavailable' with a 'Retry-After' header
> seems like the least bad option to me.
>
> Thanks,
> Nandor
>
> vignesh a <im...@gmail.com> ezt írta (időpont: 2026. júl. 16.,
> Cs, 16:44):
> >
> > Hi all,
> >
> > Following the recent [DISCUSS] thread [1], I'd like to call a vote on the
> > HTTP status code to return when a table or view rename operation fails
> with
> > TARGET_ENTITY_CONCURRENTLY_MODIFIED.
> >
> > *Background* 409 Conflict and 429 Too Many Requests were discussed as
> > alternatives. However:
> >
> > - 409 conflicts with the Iceberg REST spec's existing meaning ("target
> > identifier already exists"), so clients would misinterpret it.
> > - 429 implies rate limiting and carries Retry-After semantics that
> don't
> > apply here.
> >
> > The discussion converged on *503 Service Unavailable* as the cleanest
> > option - it is retryable and avoids the semantic conflicts above.
> >
> > *Proposal * Return HTTP *503 Service Unavailable* when a table/view
> rename
> > fails with TARGET_ENTITY_CONCURRENTLY_MODIFIED. This signals a temporary
> > condition that clients can retry. Server-side automatic retries are *out
> of
> > scope* for this change and can be considered separately later.
> >
> > My vote: *+1*
> >
> > Please cast your vote by replying to this thread:
> >
> > [ ] *+1 / Yes* -- Approve returning 503 [ ] *-1 / No* -- Reject this
> > proposal and resume discussion [ ] *0* -- Abstain
> >
> > This vote will remain open for 72 hours and close at *14:00 UTC on
> Sunday,
> > 19 July 2026*.
> >
> > Thanks, *Vignesh*
> >
> > [1] https://lists.apache.org/thread/tr8zh8121t2jb41s0q2yd9s73y2tp2tq
> >
> > This change is implemented in PR #4646
> > <https://github.com/apache/polaris/pull/4646> .
>
>
>
> --
> Kollár Nándor
>
Re: [VOTE] Return 503 Service Unavailable for TARGET_ENTITY_CONCURRENTLY_MODIFIED during table/view rename
Posted by Nándor Kollár <ko...@gmail.com>.
+1 (non-binding)
I think that '503 Service Unavailable' with a 'Retry-After' header
seems like the least bad option to me.
Thanks,
Nandor
vignesh a <im...@gmail.com> ezt írta (időpont: 2026. júl. 16.,
Cs, 16:44):
>
> Hi all,
>
> Following the recent [DISCUSS] thread [1], I'd like to call a vote on the
> HTTP status code to return when a table or view rename operation fails with
> TARGET_ENTITY_CONCURRENTLY_MODIFIED.
>
> *Background* 409 Conflict and 429 Too Many Requests were discussed as
> alternatives. However:
>
> - 409 conflicts with the Iceberg REST spec's existing meaning ("target
> identifier already exists"), so clients would misinterpret it.
> - 429 implies rate limiting and carries Retry-After semantics that don't
> apply here.
>
> The discussion converged on *503 Service Unavailable* as the cleanest
> option - it is retryable and avoids the semantic conflicts above.
>
> *Proposal * Return HTTP *503 Service Unavailable* when a table/view rename
> fails with TARGET_ENTITY_CONCURRENTLY_MODIFIED. This signals a temporary
> condition that clients can retry. Server-side automatic retries are *out of
> scope* for this change and can be considered separately later.
>
> My vote: *+1*
>
> Please cast your vote by replying to this thread:
>
> [ ] *+1 / Yes* -- Approve returning 503 [ ] *-1 / No* -- Reject this
> proposal and resume discussion [ ] *0* -- Abstain
>
> This vote will remain open for 72 hours and close at *14:00 UTC on Sunday,
> 19 July 2026*.
>
> Thanks, *Vignesh*
>
> [1] https://lists.apache.org/thread/tr8zh8121t2jb41s0q2yd9s73y2tp2tq
>
> This change is implemented in PR #4646
> <https://github.com/apache/polaris/pull/4646> .
--
Kollár Nándor
Re: [VOTE] Return 503 Service Unavailable for TARGET_ENTITY_CONCURRENTLY_MODIFIED during table/view rename
Posted by Dmitri Bourlatchkov <di...@apache.org>.
+1 (binding)
The 503 response + Retry-After header seems reasonable for this case, per
the discussion thread.
Cheers,
Dmitri.
On Thu, Jul 16, 2026 at 10:44 AM vignesh a <im...@gmail.com> wrote:
> Hi all,
>
> Following the recent [DISCUSS] thread [1], I'd like to call a vote on the
> HTTP status code to return when a table or view rename operation fails with
> TARGET_ENTITY_CONCURRENTLY_MODIFIED.
>
> *Background* 409 Conflict and 429 Too Many Requests were discussed as
> alternatives. However:
>
> - 409 conflicts with the Iceberg REST spec's existing meaning ("target
> identifier already exists"), so clients would misinterpret it.
> - 429 implies rate limiting and carries Retry-After semantics that don't
> apply here.
>
> The discussion converged on *503 Service Unavailable* as the cleanest
> option - it is retryable and avoids the semantic conflicts above.
>
> *Proposal * Return HTTP *503 Service Unavailable* when a table/view rename
> fails with TARGET_ENTITY_CONCURRENTLY_MODIFIED. This signals a temporary
> condition that clients can retry. Server-side automatic retries are *out of
> scope* for this change and can be considered separately later.
>
> My vote: *+1*
>
> Please cast your vote by replying to this thread:
>
> [ ] *+1 / Yes* -- Approve returning 503 [ ] *-1 / No* -- Reject this
> proposal and resume discussion [ ] *0* -- Abstain
>
> This vote will remain open for 72 hours and close at *14:00 UTC on Sunday,
> 19 July 2026*.
>
> Thanks, *Vignesh*
>
> [1] https://lists.apache.org/thread/tr8zh8121t2jb41s0q2yd9s73y2tp2tq
>
> This change is implemented in PR #4646
> <https://github.com/apache/polaris/pull/4646> .
>