exatn::getService failure

[DmitryLyakh]

Branch devel. After gluing together exatn::numerics and exatn::runtime it looks like CPP microservices exatn::getService does not work (it does not discover tensor runtime services). Three test fail with the same error.

[amccaskey]

I observed a failure in TAProlParserTester. It was missing exatn::initialize(). Can you post the other errors you saw, ctest --output-on-failure

[DmitryLyakh]

Test project /home/dima/src/exatn/build
Start 1: ServiceRegistryTester
1/5 Test #1: ServiceRegistryTester ............ Passed 0.00 sec
Start 2: NumericsTester
2/5 Test #2: NumericsTester ...................***Exception: SegFault 0.11 sec
[==========] Running 2 tests from 1 test case.
[----------] Global test environment set-up.
[----------] 2 tests from NumericsTester
[ RUN ] NumericsTester.checkSimple
2 1 13 0
{1:5,0:13}
2 61 32
{61,32}
{1:4}
[ OK ] NumericsTester.checkSimple (0 ms)
[ RUN ] NumericsTester.checkNumServer
Exatn not initialized before use. Please execute exatn::Initialize() before using API.
Could not find service with name eager-dag-executor. Perhaps the service is not Identifiable.
Invalid exatn Service. Could not find eager-dag-executor in Service Registry.
Exatn not initialized before use. Please execute exatn::Initialize() before using API.
Could not find service with name talsh-node-executor. Perhaps the service is not Identifiable.
Invalid exatn Service. Could not find talsh-node-executor in Service Registry.

Start 3: TAProLInterpreterTester

3/5 Test #3: TAProLInterpreterTester ..........***Exception: SegFault 0.10 sec
[==========] Running 1 test from 1 test case.
[----------] Global test environment set-up.
[----------] 1 test from TAProLInterpreterTester
[ RUN ] TAProLInterpreterTester.checkSimple
Exatn not initialized before use. Please execute exatn::Initialize() before using API.

Start 4: DirectedBoostGraphTester

4/5 Test #4: DirectedBoostGraphTester ......... Passed 0.00 sec
Start 5: TensorRuntimeTester
5/5 Test #5: TensorRuntimeTester ..............***Exception: SegFault 0.10 sec
Exatn not initialized before use. Please execute exatn::Initialize() before using API.
Could not find service with name eager-dag-executor. Perhaps the service is not Identifiable.
Invalid exatn Service. Could not find eager-dag-executor in Service Registry.
Exatn not initialized before use. Please execute exatn::Initialize() before using API.
Could not find service with name talsh-node-executor. Perhaps the service is not Identifiable.
Invalid exatn Service. Could not find talsh-node-executor in Service Registry.

40% tests passed, 3 tests failed out of 5

Total Test time (real) = 0.32 sec

The following tests FAILED:
2 - NumericsTester (SEGFAULT)
3 - TAProLInterpreterTester (SEGFAULT)
5 - TensorRuntimeTester (SEGFAULT)
Errors while running CTest

[amccaskey]

Oh cool, these are easy to fix. You just need to update the failing testers main() functions to first run exatn::initialize().

[DmitryLyakh]

I actually did that in NumericsTester but it still fails for some reason ...

[DmitryLyakh]

Oh, you meant I should put exatn::initialize() into main() instead of test itself?

[amccaskey]

Well either would do if there is only one TEST(). Did you run make install?

[DmitryLyakh]

NumericsTester has two tests and I put exatn::initialize() and exatn::finalize() in the second one since it was failing. So, I guess I should just move this to main() instead. Will try.

[DmitryLyakh]

Do you think getService not being able to find registered exatn::runtime services was because of this as well?

[amccaskey]

yes, if you don't call initialize() then the ServiceRegistry is not instantiated and initialized. ServiceRegistry::initialize() searches the ~/.exatn/plugins directory for shared libraries, loads them, and calls *Activator::start() on each one. Without doing this, there are no services available to get.

[DmitryLyakh]

Nope, it does not help either, same thing:
Exatn not initialized before use. Please execute exatn::Initialize() before using API.
Could not find service with name eager-dag-executor. Perhaps the service is not Identifiable.
Invalid exatn Service. Could not find eager-dag-executor in Service Registry.
Exatn not initialized before use. Please execute exatn::Initialize() before using API.
Could not find service with name talsh-node-executor. Perhaps the service is not Identifiable.
Invalid exatn Service. Could not find talsh-node-executor in Service Registry.

[DmitryLyakh]

Also note that it only complains about exatn runtime services: EagerGraphExecutor and TalshNodeExecutor.

[amccaskey]

This is interesting. I was able to reproduce this. It is happening because TensorRuntime::TensorRuntime(..) constructor is getting reference to the executor services before the exatn::initialize() call is finished. So exatn::initalize() gets called, which loads the bundles, which loads the MPIRPCActivator.start(), which constructs MPIServer, which calls DriverServer super constructor, which instantiates a TAProlInterpreter, which creates the NumServer, which creates the TensorRuntime, which calls getService()... all before initialize() finishes. Phew....

Simple fix is to move the Taprol parser construction from DriverServer constructor initializer list, to the MPIServer.start() method.