You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We had some very rare OOMs in simulation for a while, which are very hard to debug. In the past, I had some evidence these might be related to TraceEvent, but I don't have enough proof. This is because most of the memory used in FDB is allocated via FastAllocator, which can't be tracked.
With #2217, for the first time, I could profile the whole memory usage including FastAllocator. What I did is:
We had some very rare OOMs in simulation for a while, which are very hard to debug. In the past, I had some evidence these might be related to TraceEvent, but I don't have enough proof. This is because most of the memory used in FDB is allocated via FastAllocator, which can't be tracked.
With #2217, for the first time, I could profile the whole memory usage including
FastAllocator
. What I did is:yum install -y gperftools-devel gperftools-libs gperftools
cmake -DUSE_GPERFTOOLS=1 ../foundationdb -G Ninja; ninja
HEAPPROFILE=/tmp/fdbserver fdbserver [args...]
pprof-symbolize gperf-build/bin/fdbserver fdbserver.0065.heap
Finally, I got the profile, which can be downloaded here. The profile shows that 5914MB are allocated for TraceEvent (the total usage is 6318MB).
The text was updated successfully, but these errors were encountered: