logictest: add a kvtrace test type to logic tests. #45563
Conversation
|
This change is |
|
Oh this is pretty slick -- is it possible to do something like To only see the requested operations? Also, would it be too special cased to have the command accept a table id or an index id to filter on? |
pkg/sql/logictest/logic.go
Outdated
| return err | ||
| } | ||
| query.colTypes = "T" | ||
| query.sql = `SELECT message FROM [SHOW KV TRACE FOR SESSION] WITH ORDINALITY |
There was a problem hiding this comment.
Why do you need WITH ORDINALITY here?
There was a problem hiding this comment.
I don't know, I just cargo culted it from somewhere else :) I'll get rid of it.
|
Might also be worth it to make an issue tracking transferring all our existing tests that do the boilerplate to use this |
It's common to want to test "kv trace" output in execbuilder tests, to make sure that a particular SQL statement emits the expected set of key reads, writes, deletes, and so on. It's very annoying to re-type out the setup for kv tracing every time (it's multiple boiler plate-y lines), so this commit introduces a shortcut for this common operation. It looks like this: query T kvtrace SELECT * FROM t ---- <expected kv trace output> Release note: None
jordanlewis
force-pushed
the
kvtrace-logictest
branch
from
73ec370 to
7f6afc1
Compare
|
As far as enhancing the test type to include expected op types, I think it's a little too annoying at the moment given the way the logic test parser works. I could do it later maybe but for now I don't really want to bother, since it'll probably work for most use cases at the moment just fine. |
|
I agree it's a good idea to transfer the existing tests to use this but yeah I definitely don't want to do this right now. |
|
Sweet, both of those seem reasonable to me. code LGTM |
|
bors r+ |
Build succeeded |
45140: sql: use CommandResult for displaying ParameterStatus updates r=andreimatei a=otan We were previously sending ParameterStatus using a listener. This is refactored to talk using CommandResult instead, with results only being buffered if "Close()" is called. To do this, a new `AppendParamStatusUpdate` function to `RestrictedCommandResult`. Since setting vars live in `sessionDataMutator`, we add a new `paramStatusUpdater` interface which uses the `RestrictedCommandResult`. This var is reset each time we execute with an openTxn with the CommandResult used for it. Also renamed "StatusParam" to "ParamStatus", which is closer to the name "ParameterStatus". Release note: None 45530: ui: network no connection r=dhartunian a=elkmaster removed trigger when all the nodes are connected Resolves: #45406 Release note (ui): none 45564: sql: inverted index fixes r=jordanlewis a=jordanlewis This is based on #45563, and split out from #45157. Closes #45154. Closes #32468. Miscellaneous fixes for inverted indexes. 1. Don't produce duplicate keys for inverted indexes. 2. Permit rows to not emit any keys for a particular index, which is required to allow not having any inverted index entries for NULL (or empty container). 3. Don't produce keys for NULL entries in inverted indexes Co-authored-by: Oliver Tan <[email protected]> Co-authored-by: Vlad Los <[email protected]> Co-authored-by: Jordan Lewis <[email protected]>
It's common to want to test "kv trace" output in execbuilder tests, to
make sure that a particular SQL statement emits the expected set of key
reads, writes, deletes, and so on. It's very annoying to re-type out the
setup for kv tracing every time (it's multiple boiler plate-y lines), so
this commit introduces a shortcut for this common operation.
It looks like this:
Release note: None