[ https://issues.apache.org/jira/browse/BEAM-3060?focusedWorklogId=84387&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-84387 ]
ASF GitHub Bot logged work on BEAM-3060: ---------------------------------------- Author: ASF GitHub Bot Created on: 26/Mar/18 15:16 Start Date: 26/Mar/18 15:16 Worklog Time Spent: 10m Work Description: szewi commented on issue #4861: [BEAM-3060] Jenkins configuration allowing to run FilebasedIO tests on HDFS. URL: https://github.com/apache/beam/pull/4861#issuecomment-376202797 Hi Cham, it's ready for review. The kubernetes clusters are being created and deleted afterwards, even during failure. I trigger two builds to run in the same time and there were no interferences between them. Tests are working smoothly on HDFS for TextIO, CompressedText, AVRO and XMLIO. There is only an issue for TFRecord: https://builds.apache.org/job/beam_PerformanceTests_TFRecordIOIT_HDFS/3/console >java.lang.IllegalStateException: Not a valid TFRecord. Fewer than 12 bytes. >java.lang.IllegalStateException: Invalid data this is only happening when running TFRecord tests on HDFS. Same happen when running tests from my local machine. This seem to be related to how data is accessed in TFRecord tests https://github.com/apache/beam/blob/597e3f92bc8be692d5d8e8040b33ce0c77350fa2/sdks/java/io/file-based-io-tests/src/test/java/org/apache/beam/sdk/io/tfrecord/TFRecordIOIT.java#L110 and https://github.com/apache/beam/blob/597e3f92bc8be692d5d8e8040b33ce0c77350fa2/sdks/java/io/file-based-io-tests/src/test/java/org/apache/beam/sdk/io/tfrecord/TFRecordIOIT.java#L113 I investigated it and it seems like even if I specify exact filename on hdfs (`.apply(TFRecordIO.read().from("hdfs://35.225.39.200:9000/TFRecord_1522073710252-00000-of-00006.tfrecord").withCompression(AUTO))`) instead of filenamePattern it still got the same error with invalid data. I can create a Bug in JIRA describing the issue, providing steps to reproduce and look at this separately. Should we proceed with that PR, but simply temporarily disable/remove TFRecord job? WDYT? ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org Issue Time Tracking ------------------- Worklog Id: (was: 84387) Time Spent: 6h 20m (was: 6h 10m) > Add performance tests for commonly used file-based I/O PTransforms > ------------------------------------------------------------------ > > Key: BEAM-3060 > URL: https://issues.apache.org/jira/browse/BEAM-3060 > Project: Beam > Issue Type: Test > Components: sdk-java-core > Reporter: Chamikara Jayalath > Assignee: Szymon Nieradka > Priority: Major > Time Spent: 6h 20m > Remaining Estimate: 0h > > We recently added a performance testing framework [1] that can be used to do > following. > (1) Execute Beam tests using PerfkitBenchmarker > (2) Manage Kubernetes-based deployments of data stores. > (3) Easily publish benchmark results. > I think it will be useful to add performance tests for commonly used > file-based I/O PTransforms using this framework. I suggest looking into > following formats initially. > (1) AvroIO > (2) TextIO > (3) Compressed text using TextIO > (4) TFRecordIO > It should be possibly to run these tests for various Beam runners (Direct, > Dataflow, Flink, Spark, etc.) and file-systems (GCS, local, HDFS, etc.) > easily. > In the initial version, tests can be made manually triggerable for PRs > through Jenkins. Later, we could make some of these tests run periodically > and publish benchmark results (to BigQuery) through PerfkitBenchmarker. > [1] https://beam.apache.org/documentation/io/testing/ -- This message was sent by Atlassian JIRA (v7.6.3#76005)