RFCs: propose a syntax for pausing/resuming/canceling operations

[benesch]

No description provided.

[rjnn]

This change is 

[petermattis]

Review status: 0 of 1 files reviewed at latest revision, 1 unresolved discussion, all commit checks successful.

---

docs/RFCS/pause_resume_cancel.md, line 92 at r1 (raw file):

```sql
ALTER SYSTEM KILL SESSION <session-id>;

ALTER JOB [PAUSE|RESUME|CANCEL] seems more natural to me, though I'm not opposed to adding new top-level syntax.

---

Comments from Reviewable

[bdarnell]

I'm fine with either the proposed syntax or Peter's ALTER (QUERY|JOB) (CANCEL|PAUSE|RESUME)

---

Review status: 0 of 1 files reviewed at latest revision, 3 unresolved discussions, all commit checks successful.

---

docs/RFCS/pause_resume_cancel.md, line 71 at r1 (raw file):

rolls back all backfill progress.

## Postgres syntax

Postgres additionally has support for cancellation built in to the wire protocol: send a special packet on a new TCP connection (bypassing all the regular authentication. in its place, you pass a token that the server sends at the start of each connection).

I have no idea how common each mode of operation is in practice. My guess is that only very postgres-specific tools like psql use the protocol-based mode, and anything more generic would use the function.

---

docs/RFCS/pause_resume_cancel.md, line 84 at r1 (raw file):

seem like they *should* have new syntax.

Alas, we may eventually end up supporting this syntax for compatibility's sake.

What type did we end up using as our query ids? Integers or bytes/strings? We may only be able to support pg_cancel_backend if we use integer query ids.

---

Comments from Reviewable

[itsbilal]

---

Review status: 0 of 1 files reviewed at latest revision, 3 unresolved discussions, all commit checks successful.

---

docs/RFCS/pause_resume_cancel.md, line 84 at r1 (raw file):

Previously, bdarnell (Ben Darnell) wrote…

What type did we end up using as our query ids? Integers or bytes/strings? We may only be able to support pg_cancel_backend if we use integer query ids.

We haven't decided on that yet, but good point - we may want to stick to integer counters as query IDs for this reason alone. We can also discuss this around the time my query cancellation RFC is out for review; it doesn't have to be finalized for this RFC.

---

Comments from Reviewable

[benesch]

IDs generated with unique_rowid or its moral equivalent? We should check to see if Postgres IDs are int8s and not int4s.

…

On Thu, Jun 1, 2017 at 5:51 PM Bilal Akhtar ***@***.***> wrote: [image: <img class="emoji" title=":lgtm:" alt=":lgtm:" align="absmiddle" src="https://reviewable.io/lgtm.png" height="20" width="61"/>] <https://camo.githubusercontent.com/41b3ec3419116e3b3f84507ad965646ce4ce1f0c/68747470733a2f2f72657669657761626c652e696f2f6c67746d2e706e67> ------------------------------ Review status: 0 of 1 files reviewed at latest revision, 3 unresolved discussions, all commit checks successful. ------------------------------ *docs/RFCS/pause_resume_cancel.md, line 84 at r1 <https://reviewable.io:443/reviews/cockroachdb/cockroach/16273#-Kl__TrpQIf3u75VthcP:-Kl_bqUAeWt-qqr-8cyG:bu2uh6e> (raw file <https://github.com/cockroachdb/cockroach/blob/c005d596c79172c1a809689c88bd9a82e636acfd/docs/RFCS/pause_resume_cancel.md#L84>):* *Previously, bdarnell (Ben Darnell) wrote…* What type did we end up using as our query ids? Integers or bytes/strings? We may only be able to support pg_cancel_backend if we use integer query ids. We haven't decided on that yet, but good point - we may want to stick to integer counters as query IDs for this reason alone. We can also discuss this around the time my query cancellation RFC is out for review; it doesn't have to be finalized for this RFC. ------------------------------ *Comments from Reviewable <https://reviewable.io:443/reviews/cockroachdb/cockroach/16273#-:-Kl_ctVXj3U1A6wgNYdc:bnfp4nl>* — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#16273 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA15IHzrOrFTBHw3gAGjYsfrJghB1MeNks5r_zJFgaJpZM4NtXP2> .

[bdarnell]

Postgres "backend IDs" used for cancellation are 32 bits.

[a-robinson]

Thanks for including how different systems approach the problem!

---

Reviewed 1 of 1 files at r1.
Review status: all files reviewed at latest revision, 3 unresolved discussions, all commit checks successful.

---

Comments from Reviewable

[jordanlewis]

IMO, CANCEL/PAUSE/RESUME is more natural than ALTER QUERY/JOB.

---

Review status: all files reviewed at latest revision, 3 unresolved discussions, all commit checks successful.

---

Comments from Reviewable

[benesch]

Since it seems no one is opposed to CANCEL/PAUSE/RESUME and we have a few votes in favor of it: entering final comment period!

[vivekmenezes]

---

Review status: all files reviewed at latest revision, 3 unresolved discussions, all commit checks successful.

---

Comments from Reviewable

[benesch]

Hearing no objections, I'm merging!