re-enable recovery test after rabit allow duplicated allreduce calls with same signature

[chenqin]

Reeanble rabit tests after fix
dmlc/rabit#128 merged

Rabit assume bootstrap allreduce has unique signature, in #5012 we find it failed to satisfy such constraint.

The goal of this pr is to loosen such constraint and allow last write overwrite results with same signature during bootstrap phase. Here is example https://github.com/dmlc/xgboost/pull/4990/files#diff-5fabd019d3f1af7572baa4a6301cf076R36

[trivialfis]

Restarted the aborted tests.

[codecov-io]

Codecov Report

❗ No coverage uploaded for pull request base (master@1733c9e). Click here to learn what that means.
The diff coverage is n/a.

@@            Coverage Diff            @@
##             master    #5031   +/-   ##
=========================================
  Coverage          ?   71.52%           
=========================================
  Files             ?       11           
  Lines             ?     2311           
  Branches          ?        0           
=========================================
  Hits              ?     1653           
  Misses            ?      658           
  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 1733c9e...5dd5a75. Read the comment docs.

[chenqin]

#5033

[chenqin]

@trivialfis can we rerun tests, a bit weird.

[hcho3]

The error message in the R tests is

undefined symbol: _ZN5rabit6engine9AllgatherEPvmmmm

I thought the R package was using mock Rabit? (Distributed training is not supported for XGBoost-R)

[chenqin]

The error message in the R tests is

undefined symbol: _ZN5rabit6engine9AllgatherEPvmmmm

I thought the R package was using mock Rabit? (Distributed training is not supported for XGBoost-R)

I see, that’s probably not related to mock but all gather interface merged in master.

[chenqin]

depends on dmlc/rabit#129
@trivialfis

[trivialfis]

Will look into this once our Windows worker is back. Sorry for being slow here.

[hcho3]

@trivialfis Sorry for the delay, I was spending time with my family over the Thanksgiving holiday. The blocking issue now is inability to run the compiled C++ tests (testxgboost.exe) in Windows workers; the executable simply crashes. I'm investigating now.

[hcho3]

@trivialfis I found the reason why testxgboost.exe was crashing. I had to install a system lib called Microsoft Visual C++ Redistributable for Visual Studio. #5061 will be resolved shortly.

[trivialfis]

@hcho3. That's great!

[trivialfis]

@chenqin Could you make a rebase onto master branch. As you used your master branch I can't do that for you.

[trivialfis]

@chenqin pls help rebasing, otherwise the Jenkins Windows tests can not pass.

[chenqin]

@nateagr
dmlc/rabit#113

`
Error: package or namespace load failed for 'xgboost' in dyn.load(file, DLLpath = DLLpath, ...):

[2019-12-15T01:52:26.359Z] unable to load shared object '/workspace/xgboost/xgboost.Rcheck/xgboost/libs/xgboost.so':

[2019-12-15T01:52:26.359Z] /workspace/xgboost/xgboost.Rcheck/xgboost/libs/xgboost.so: undefined symbol: ZN5rabit6engine9AllgatherEPvmmmmPKciS3
`

[chenqin]

@trivialfis @hcho3 @CodingCat happy new year

[hcho3]

@chenqin You too

[trivialfis]

@CodingCat Could you please share the rabit bug you discovered? My next target would be merging existing PRs. No more new bug fix from me unless critical.

[CodingCat]

Why we stop bug fixing but merging new functionalities? It’s counter-intuitive as the final steps of releasing a new version?

…

On Wed, Jan 1, 2020 at 11:34 PM Jiaming Yuan ***@***.***> wrote: @CodingCat <https://github.com/CodingCat> Could you please share the rabit bug you discovered? My next target would be merging PRs. No more new bug fixes unless critical. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#5031?email_source=notifications&email_token=AAFFQ6GAJCKYMU7YNWWL3STQ3WKIFA5CNFSM4JL75I7KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEH5ZIZA#issuecomment-570135652>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFFQ6BLT5TH6EPEY6PKFWTQ3WKIFANCNFSM4JL75I7A> .

[trivialfis]

I mean I won't focus on bug fixes, others can continue the process. And merging PRs will focus on bug fixes PRs.

[trivialfis]

Clarified, sorry for the ambiguity.

[chenqin]

thanks @trivialfis helping get enable rabit test pr in. close this pr