Skip to content

Conversation

@carlopi
Copy link
Contributor

@carlopi carlopi commented May 18, 2025

This restore the following workflows to work as intended:

FORCE INSTALL spatial;
--- not relevant which extension, this should trigger a full download
UPDATE EXTENSIONS (spatial);
--- this should be a single GET request with "If-Not-Match" header

and the same with httpfs loaded:

FORCE INSTALL httpfs;
LOAD httpfs;
FORCE INSTALL spatial;
--- not relevant which extension, this should trigger a full download
UPDATE EXTENSIONS (spatial);
--- this should be a single GET request with "If-Not-Match" header

Note that this is relevant only for UPDATE EXTENSIONS or for the INSTALL codepath for an already installed extension.

Also, without this PR the behaviour would be long waits on UPDATE EXTENSIONS, that would eventually timeout and resume normal execution.

It's not completely clear to me why both follow_location and keep_alive needs to be set to false, whether this has something to do with our current setup to serve extensions or else, but this restore to the defaults in place before the HTTPUtil rework.

@carlopi
Copy link
Contributor Author

carlopi commented May 18, 2025

Testing I think should happen via https://github.com/carlopi/duckdb/actions/runs/15093857243 / extension update workflow (that is currently failing). I will review that once that's in.

@carlopi
Copy link
Contributor Author

carlopi commented May 18, 2025

Failure looks to me to be not-connected, but would be handy to restart to double check since both are in the area of networking.

To me Python one looks to be random network failure, and SELECT request.type, parse_filename(request.url) FROM duckdb_logs_parsed('HTTP'); logs one to be possibly non-deterministic?
Unsure.

@Tishj
Copy link
Contributor

Tishj commented May 18, 2025

https://github.com/duckdb/duckdb/actions/runs/15093832916/job/42426425248?pr=17527 looks to be a missing ORDER BY in that test.
(probably on timestamp)

@carlopi
Copy link
Contributor Author

carlopi commented May 18, 2025

https://github.com/duckdb/duckdb/actions/runs/15093832916/job/42426425248?pr=17527 looks to be a missing ORDER BY in that test. (probably on timestamp)

I think it's the macro that should probably perform the ORDER BY, or at least I would assume logs to be ordered.

@Mytherin Mytherin merged commit 3c8f023 into duckdb:v1.3-ossivalis May 19, 2025
48 of 50 checks passed
@Mytherin
Copy link
Collaborator

Thanks!

@carlopi carlopi deleted the fix_update_extensions branch May 19, 2025 07:15
carlopi added a commit to carlopi/duckdb_httpfs that referenced this pull request May 19, 2025
carlopi added a commit to carlopi/duckdb_httpfs that referenced this pull request May 19, 2025
carlopi added a commit to carlopi/duckdb_httpfs that referenced this pull request May 19, 2025
Mytherin added a commit to duckdb/duckdb-httpfs that referenced this pull request May 19, 2025
carlopi added a commit to carlopi/duckdb that referenced this pull request May 20, 2025
This includes both removing patch from duckdb#17527
and fix from duckdb/duckdb-httpfs#57
Mytherin added a commit that referenced this pull request May 21, 2025
This includes both removing patch from
#17527 and fix from
duckdb/duckdb-httpfs#57
krlmlr added a commit to duckdb/duckdb-r that referenced this pull request May 21, 2025
Fix update extensions (duckdb/duckdb#17527)
[Python Dev] Using `reinterpret_steal` breaks the refcount of the passed-in object (duckdb/duckdb#17525)
krlmlr added a commit to duckdb/duckdb-r that referenced this pull request May 21, 2025
Fix update extensions (duckdb/duckdb#17527)
[Python Dev] Using `reinterpret_steal` breaks the refcount of the passed-in object (duckdb/duckdb#17525)
krlmlr added a commit to duckdb/duckdb-r that referenced this pull request May 21, 2025
Fix update extensions (duckdb/duckdb#17527)
[Python Dev] Using `reinterpret_steal` breaks the refcount of the passed-in object (duckdb/duckdb#17525)
krlmlr added a commit to duckdb/duckdb-r that referenced this pull request May 23, 2025
Fix update extensions (duckdb/duckdb#17527)
[Python Dev] Using `reinterpret_steal` breaks the refcount of the passed-in object (duckdb/duckdb#17525)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants