Intermittent travis failures due to potential_deadlock_detected #6515

issue laanwj opened this issue on August 4, 2015

laanwj commented at 2:10 PM on August 4, 2015: member

Running 161 test cases...
test_bitcoin: sync.cpp:112: void potential_deadlock_detected(const std::pair<void*, void*>&, const LockStack&, const LockStack&): Assertion `onlyMaybeDeadlock' failed.
unknown location(0): fatal error in "rpc_wallet": signal: SIGABRT (application abort requested)
test/rpc_wallet_tests.cpp(64): last checkpoint
test_bitcoin: key.cpp:198: void ECC_Start(): Assertion `secp256k1_context == __null' failed.
unknown location(0): fatal error in "script_CHECKMULTISIG23": signal: SIGABRT (application abort requested)
test/rpc_wallet_tests.cpp(64): last checkpoint
test_bitcoin: key.cpp:198: void ECC_Start(): Assertion `secp256k1_context == __null' failed.
unknown location(0): fatal error in "script_valid": signal: SIGABRT (application abort requested)
test/rpc_wallet_tests.cpp(64): last checkpoint
test_bitcoin: key.cpp:198: void ECC_Start(): Assertion `secp256k1_context == __null' failed.
unknown location(0): fatal error in "script_PushData": signal: SIGABRT (application abort requested)
test/rpc_wallet_tests.cpp(64): last checkpoint

laanwj added the label Build system on Aug 4, 2015
casey commented at 4:26 PM on August 4, 2015: contributor

For reference, here are some travis builds that failed:

https://travis-ci.org/bitcoin/bitcoin/builds/73939528 https://travis-ci.org/bitcoin/bitcoin/builds/73941114 https://travis-ci.org/bitcoin/bitcoin/builds/73943358 https://travis-ci.org/bitcoin/bitcoin/builds/73953014

All the changes look pretty harmless. And I can't reproduce it locally. It would be nice to have the debug log from travis runs, but that doesn't seem to be available.
laanwj renamed this:
~~Intermittend travis failures due to potential_deadlock_detected~~
Intermittent travis failures due to potential_deadlock_detected
on Aug 4, 2015
TheBlueMatt commented at 9:25 PM on August 4, 2015: member

Probably a good idea to print the deadlock error to stderr instead of debug.log to make these kinds of issues easier to debug.
casey commented at 9:31 PM on August 4, 2015: contributor
Do you happen to know where debug.log is on the travis build?

We could add the following to .travis.yml:
```
after_failure:
   - cat /home/travis/.bitcoin/debug.log
```
And the logs would be printed after every test failure. I'm not sure that's where debug.log will wind up though, only guessing.
theuni commented at 2:04 AM on August 6, 2015: member

This is easy to reproduce locally via ./configure --enable-debug. I haven't looked into it yet, but it appears to be a static initialization issue, likely some uninteresting change upset the init order. I'll dive in tomorrow.
laanwj commented at 12:22 PM on August 6, 2015: member

Almost all Travis runs were failing. I've disabled a4fe57da6207c1e5691a1e843d22db571f3f0186 for now.
laanwj closed this on Aug 10, 2015
MarcoFalke locked this on Sep 8, 2021

Contributors

Labels

Linked (view graph)

#6534 Fix test locking issues and un-revert the probable-deadlines assertions commit