Convert _insert_graph_receipts_txn to simple_upsert

[realtyem]

Which should cut down on Postgres log spam like:

2023-09-11 05:37:47.811 CDT [13934] ERROR:  could not serialize access due to concurrent update
2023-09-11 05:37:47.811 CDT [13934] STATEMENT:  INSERT INTO receipts_graph (room_id, receipt_type, user_id, thread_id, event_ids, data) VALUES ('!room_1:computer.surgery', 'm.read', '@some_redacted_user:gnome.org', 'main', '["$hCBJpUJwHjjG6ztFwHrDPtqHFyMQ8eWMOCRUhmO3UBs"]', '{"thread_id":"main","ts":1694427550626}') ON CONFLICT (room_id, receipt_type, user_id, thread_id)  DO UPDATE SET event_ids=EXCLUDED.event_ids, data=EXCLUDED.data
2023-09-11 05:52:03.258 CDT [13934] ERROR:  could not serialize access due to concurrent update
2023-09-11 05:52:03.258 CDT [13934] STATEMENT:  INSERT INTO receipts_graph (room_id, receipt_type, user_id, thread_id, event_ids, data) VALUES ('!alCakyySsFIAVfZLDL:matrix.org', 'm.read', '@some_other_redacted_user:personal_homeserver.uk', 'main', '["$ekmtqEQ5PFNHFg49-zZamgw6LDNcwqYMh0dyGSQ3A2E"]', '{"thread_id":"main","ts":1694429513585}') ON CONFLICT (room_id, receipt_type, user_id, thread_id)  DO UPDATE SET event_ids=EXCLUDED.event_ids, data=EXCLUDED.data
2023-09-11 05:57:43.559 CDT [14238] ERROR:  could not serialize access due to concurrent update
2023-09-11 05:57:43.559 CDT [14238] STATEMENT:  INSERT INTO receipts_graph (room_id, receipt_type, user_id, thread_id, event_ids, data) VALUES ('!no_idea_room_id:matrix.org', 'm.read', '@another_random_user_id:other_homeserver.pt', 'main', '["$nBSxdTs9h73YOLm40A-ggrfSOvv7KL-iJHOaRzOjMnY"]', '{"thread_id":"main","ts":1694429518348}') ON CONFLICT (room_id, receipt_type, user_id, thread_id)  DO UPDATE SET event_ids=EXCLUDED.event_ids, data=EXCLUDED.data

Pull Request Checklist

• Pull request is based on the develop branch
• Pull request includes a changelog file.
• Pull request includes a sign off
• Code style is correct
  (run the linters)

Signed-off-by: Jason Little realtyem@gmail.com

[erikjohnston]

It looks like _insert_graph_receipt_txn is basically just an upsert, which we can do as an "autocommit". I'd suggest we change _insert_graph_receipt_txn to use await simple_upsert(..) instead so we get that for free.

This saves having to reason about if its "safe" or not.

[realtyem]

It looks like _insert_graph_receipt_txn is basically just an upsert, which we can do as an "autocommit". I'd suggest we change _insert_graph_receipt_txn to use await simple_upsert(..) instead so we get that for free.

This safes having to reason about if its "safe" or not.

Ooooooooo, neat idea. I did not know you could do that. Do we ever actually read this table?

[clokep]

Do we ever actually read this table?

No and I debated about removing it. It is somewhat there for "correctness" but we never implemented that. 🤷

[realtyem]

Will need to update the title and the changelog(and maybe the function name), but maybe this looks ok?

[realtyem]

Should I separate the cache invalidation to move the 'linearized' related up near it's own transaction instead of hiding it inside another unrelated function?

Edit: Actually, I think the cache for the linearized version is already invalidated in it's own transaction function, _insert_linearized_receipt_txn. This looks like it could get messy to untangle. Let's not do this here on this PR.

[erikjohnston]

Thanks!