Skip to content

GML-2200 Release 2.0.5 - #288

Open
chengbiao-jin wants to merge 6 commits into
masterfrom
release_2.0.5
Open

chengbiao-jin wants to merge 6 commits into
masterfrom
release_2.0.5

Conversation

@chengbiao-jin

@chengbiao-jin chengbiao-jin commented Sep 28, 2026 •

Copy link
Copy Markdown
Collaborator
  • Fix async connection use across event loops (GML-2183)

    • Reuse a connection after the loop it first ran on has closed
    • Keep sessions per loop so threads no longer interfere
  • Submit query installation asynchronously (GML-2201)

    • Behavior change: when not waiting, return the id of the submitted
      request instead of a completion message. This is the default for
      async connections.
  • Add chat engine and response field selection to AI queries (GML-2202)

  • Fix vertex id escaping for installed queries (GML-2203, GML-2204)

    • Percent sign in ids passed via POST
    • Ampersand or hash in untyped vertex ids passed via GET
  • Prepare the 2.0.5 release (GML-2200)

    • Behavior change: raise the minimum supported Python version to 3.9,
      from 3.8. Installing on 3.8 previously succeeded and then failed on
      import.
  • Point tests at a live instance through TG_TEST_ environment
    variables, so its address and credentials need not be written into a
    tracked config file

Verified against a live TigerGraph 4.2.2 instance: the whole suite was run
with no tests skipped. Every remaining failure also occurs on master — each
differing suite was re-run interleaved against master and matched — except
the GDS suites, which need torch and the Cora and hetero datasets that this
instance does not have. The repo runs no CI, so none of this ran automatically.

- Reuse a connection after the event loop it first ran on has closed.
- Share one connection between threads that each run their own loop,
  without one closing the session another is using.
- Release sessions belonging to a closed loop instead of leaking them.
- Installing queries no longer holds the connection open for the whole
  compile, which could be dropped by an intermediate gateway.
- Callers that choose not to wait now receive the request id to poll,
  rather than a completion message.
- Choose the agentic or classic engine, and the variant within it.
- Choose which fields the response returns beyond the answer.
- Asking a question with no options given behaves exactly as before.
- Retrieve a vertex whose id contains a percent sign.
- Retrieve an untyped vertex whose id contains an ampersand or a hash.
- Bump the version to 2.0.5.
- Correct the minimum supported Python version to 3.9; installing on
  3.8 previously succeeded and then failed on import.
- Correct the test setup documentation, which claimed the script drops
  every graph on the instance rather than only its own.
@chengbiao-jin chengbiao-jin changed the title Release 2.0.5 GML-2200 Release 2.0.5 Sep 28, 2026
- Test connection settings can be overridden with TG_TEST_<KEY>
  variables, so a real instance no longer needs its address or
  credentials written into a tracked config file.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant