Disable install Dask master in CI

[pentschev]

No description provided.

[ajschmidt8]

This will require a manual merge once CI is complete.

[ajschmidt8]

@pentschev, can you elaborate on the numba change in 9d76971?

[pentschev]

I thought the build failure was about the Numba version, because the rest of RAPIDS now pins Numba>=0.53.1:

20:40:36 /tmp/tmp26qug3g1/dask-cuda-21.06.00a-py37_56.tar.bz2: C1114 Found invalid dependency "numba >=0.50.0,!=0.51.0" in info/index.json

With the above said, I'm not 100% certain this is the (only) issue.

[pentschev]

The Numba error is now gone (or not even reached), but there's this one now:

20:52:46 TESTS FAILED: dask-cuda-21.06.00a-py37_57.tar.bz2
20:52:49     [gpuci_conda_retry] conda returned exit code: 1
20:52:49     [gpuci_conda_retry] Exiting, no retryable conda errors detected: 'ChecksumMismatchError:' or 'CondaHTTPError:' or 'JSONDecodeError:' or 'ChunkedEncodingError:' or 'CondaMultiError:'
20:52:49     [gpuci_conda_retry] 
20:52:49 Build step 'Execute shell' marked build as failure
20:52:53 [Set GitHub commit status (universal)] ERROR on repos [GHRepository@30668f7b[nodeId=MDEwOlJlcG9zaXRvcnkxNjU5MjY5Nzc=,description=Utilities for Dask and CUDA interactions,homepage=,name=dask-cuda,fork=false,archived=false,size=1100,milestones={},language=Python,commits={},source=<null>,parent=<null>,isTemplate=<null>,url=https://api.github.com/repos/rapidsai/dask-cuda,id=165926977,nodeId=<null>,createdAt=2019-01-15T21:33:16Z,updatedAt=2021-06-07T16:04:05Z]] (sha:9d76971) with context:gpuCI/dask-cuda/build/python/3.7
20:52:53 Setting commit status on GitHub for https://github.com/rapidsai/dask-cuda/commit/9d7697182f2dd4a74da8f8d2d051d44b7a1eacb1
20:52:53 Finished: FAILURE

This looks like a CI issue, could you check @ajschmidt8 ?

[ajschmidt8]

@pentschev, I'm seeing this error message hidden in the log text. Can you take a look?

[jakirkham]

@galipremsagar would you be able to look at the issue AJ shared above? 🙂

[pentschev]

I'm seeing this error message hidden in the log text. Can you take a look?

Wait a second, I'm confused now. What's this build doing exactly? This error is because we're picking Dask 2021.06.0, which is exactly what I'm trying to prevent with this PR. AFAIK, 21.06 should pin to Dask 2021.05.1.

[pentschev]

@jakirkham I think Prem is good, the problem is that we're picking Dask 2021.06.0, which is not what RAPIDS should be pinning to.

[jakirkham]

If this is related to getting the newer dask + distributed versions, we likely need to constrain them

[ajschmidt8]

If this is related to getting the newer dask + distributed versions, we likely need to constrain them

That sounds like it could be the problem. @pentschev, thoughts? Can you try updating your PR to include this?

EDIT---
whoops, I missed the above conversation by not refreshing my page. I see the latest commit now.

[codecov-commenter]

Codecov Report

❗ No coverage uploaded for pull request base (branch-21.06@0d6c3b7). Click here to learn what that means.
The diff coverage is n/a.

❗ Current head a610297 differs from pull request most recent head 745b3a9. Consider uploading reports for the commit 745b3a9 to get more accurate results

@@               Coverage Diff               @@
##             branch-21.06     #645   +/-   ##
===============================================
  Coverage                ?   90.56%           
===============================================
  Files                   ?       15           
  Lines                   ?     1622           
  Branches                ?        0           
===============================================
  Hits                    ?     1469           
  Misses                  ?      153           
  Partials                ?        0           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0d6c3b7...745b3a9. Read the comment docs.

[pentschev]

For some reason, the test that failed is supposed to be skipped in CUDA < 11.2 but it ran anyway on the 11.0 build. I'm rerunning the tests and trying to reproduce locally in the meantime.

[pentschev]

rerun tests

[jakirkham]

It's encouraging to see the other failures went away though :)

[pentschev]

The condition was wrong, I think the latest commit will fix the issue.

[pentschev]

Everything passes now! @ajschmidt8 could you merge this?

[ajschmidt8]

CI is passing. Versions match the integration repo. merging to 21.06.