Hi Taewoo, So recently we did a refactoring change to FileSplit and FileReference classes. Part of the change was that an index that is managed by asterixdb can only be inside an IO device. Looking at the error trace, seems like a new incorrect use of the FileSplit where a FileSplit is defined as an absolute file split and the path points to a place outside all IO devices, hence the IO manager throws an exception.
Looking at the file path, it seems like it is inside an IO device but could be in the wrong node? Ultimately, we should not use full paths and instead use a relative path: instead of ../asterix-server/target/tmp/asterix_nc1/iodevice1/storage/partition_0/test/MyData_idx_fulltext_index_title/2016-12-19-22-32-36-543_2016-12-19-22-32-36-543_i, we should use partition_0/test/MyData_idx_fulltext_index_title/2016-12-19-22-32-36-543_2016-12-19-22-32-36-543_i and we can select the IO device number or leave it to the IOManager to select one of the IO devices. Q. Was this triggered on master??? > On Dec 20, 2016, at 10:08 AM, Taewoo Kim <[email protected]> wrote: > > Hi all, > > I am now working on index-based version of full-text search. The full-text > index is actually SINGLE_PARTITION_WORD_INVIX type, rather than > LENGTH_PARTITIONED_WORD_INVIX type that we are currently using. I know that > we are not using this index type quite a long time. But, the strange thing > is that the build is always successful. That means, execution test succeeds > without any issue. But, for the NCServiceExecutionIT test, the following > error is generated. Is there anybody who saw this error before? Any > comments would be appreciated. > > https://asterix-jenkins.ics.uci.edu/job/asterix-gerrit-integration-tests/org.apache.asterix$asterix-server/1352/testReport/junit/org.apache.asterix.server.test/NCServiceExecutionIT/test_NCServiceExecutionTest_30__fulltext__fulltext_index_01_/ > > Best, > Taewoo
