possible data race on the non-thread-safe NGServer.allNailStats

[yilongli]

Hi, I am a developer of the K framework. While we are trying to catch the data races inside the K framework with a dynamic race detector, the race detector also prints out some race reports inside Nailgun, which K uses heavily.

Here is one:

Data race on field java.util.HashMap.$state: {{{
    Concurrent write in thread T15 (locks held: {Monitor@444ecc49})
 ---->  at com.martiansoftware.nailgun.NGServer.getOrCreateStatsFor(NGServer.java:245)
        - locked Monitor@444ecc49 at com.martiansoftware.nailgun.NGServer.getOrCreateStatsFor(NGServer.java:241) 
        at com.martiansoftware.nailgun.NGServer.nailStarted(NGServer.java:258)
        at com.martiansoftware.nailgun.NGSession.run(NGSession.java:314)
    T15 is created by T13
        at com.martiansoftware.nailgun.NGSessionPool.take(NGSessionPool.java:85)

    Concurrent read in thread T14 (locks held: {})
 ---->  at com.martiansoftware.nailgun.NGServer.nailFinished(NGServer.java:269)
        at com.martiansoftware.nailgun.NGSession.run(NGSession.java:332)
    T14 is created by T13
        at com.martiansoftware.nailgun.NGSessionPool.take(NGSessionPool.java:85)
}}}

It's basically saying that there are concurrent and conflicting accesses to NGServer.allNailStats, which is a HashMap, and that the allNailStats.get() operation is not properly synchronized. It seems the allNailStats.get() should also be wrapped inside a synchronization block.

Can you please confirm whether it is a false positive of the race detector? Thanks!

[sbalabanov-zz]

Closing all issues older than 6 months; please reopen if it is still relevant