specify port range for the testing ignite instance

[tyrasd]

This should allow unit tests to be run in parallel without interfering with each other. (hopefully 🤞 )

[joker234]

Is there a issue related to this PR? What does this PR try to solve? Parallel executions of mvn test? If yes, the solution wasn't sufficient, I still get Exceptions from the ignite test instance.

This are example errors I get when running in parallel in at least two shells in separate working directories:

% mvn test
…
[INFO] Results:
[INFO] 
[ERROR] Errors: 
[ERROR]   TestMapReduceOSHDB_Ignite_ScanQuery.testOSMContributionView:6->TestMapReduce.testOSMContributionView:58 » ComputeUserUndeclared
[ERROR]   TestMapReduceOSHDB_Ignite_ScanQuery.testOSMContributionViewStream:6->TestMapReduce.testOSMContributionViewStream:123 » Cache
[ERROR]   TestMapReduceOSHDB_Ignite_ScanQuery.testOSMEntitySnapshotView:6->TestMapReduce.testOSMEntitySnapshotView:91 » ComputeUserUndeclared
[ERROR]   TestMapReduceOSHDB_Ignite_ScanQuery.testOSMEntitySnapshotViewStream:6->TestMapReduce.testOSMEntitySnapshotViewStream:160 » Cache
[ERROR]   TestMapReduceOSHDB_Ignite_ScanQuery.testTimeoutMapReduce »  Unexpected excepti...
[INFO] 
[ERROR] Tests run: 125, Failures: 0, Errors: 5, Skipped: 0
…

[tyrasd]

Parallel executions of mvn test?

yes, that was the rationale behind it.

[tyrasd]

errors I get when running in parallel in at least two shells in separate working directories

confirming. I thought these were gone in my testing, but maybe I didn't test it properly.

The first actual exception I get is:

class org.apache.ignite.IgniteCheckedException: Failed to serialize object: org.heigit.bigspatialdata.oshdb.api.mapreducer.backend.OSHDBIgniteMapReduceComputeTask$1@[…]
[…]
Caused by: java.io.NotSerializableException: org.heigit.bigspatialdata.oshdb.index.XYGridTree$3

🤔

[tyrasd]

ok, I guess the problem was that the ipFinder property still had all possible ports listed (<value>127.0.0.1:47577..47599</value>), so the test client would always connect to the first test-server even when there were multiple ones running.

This should be addressed "properly" now (choosing a 1 in 1000 random port for each test run).