I am wondering if this isn’t a multi-threading problem. When a rollover occurs the compression action is delegated to a separate thread so the main thread can get back to logging. It may simply be that the test case is trying to check the compressed file before it has actually finished being created.
Ralph > On Feb 21, 2016, at 6:14 AM, Stefan Bodewig <[email protected]> wrote: > > Hi Remko > > On 2016-02-21, Remko Popma wrote: > >> The funny/funky thing about this issue is that it only happens on our >> (Ubuntu) Jenkins continuous integration server, not when we build locally. >> (System info: https://builds.apache.org/computer/ubuntu-4/systemInfo) > >> To make it extra fun it's sporadic and transient: It doesn't always happen, >> and it often goes away without us making any changes... > >> Any tips on how we can debug this? > > First of all we need to figure out whether this is a bug on the reading > or the writing side. For this we really need to take a look at the file > that cannot be read. My best advice is to immediatly download the file > in question from the Jenkins workspace when it happens the next time. > > The bzip2 format is a block based compression algorithm with a block > size that's a multiple of 100kB and one of the first bytes of the > compressed file indicates the block size as a number between 1 and 9 > inclusive - for the allowed block sizes of 100k up to 900k. The file in > question seems to contain a different value, at least that's what the > exception says. > > I don't know what you do in order to create the file. Faced with > sporadic failures like this I'd suspect a race condition somewhere, > which the single-threaded BZip2CompressorInputStream is less likely to > face. > > Stefan > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [email protected] > For additional commands, e-mail: [email protected] > > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
