• 24 cpus, 64 GB memory, 1 TB hard drive
• ubuntu 15.10
• mongod 3.4.0
• standalone with: --wiredTigerCacheSizeGB 8 --wiredTigerCollectionBlockCompressor=zlib

Run 20 threads of attached repro script with:

    for t in $(seq 20); do
        mongo --quiet test --eval "load('repro.js'); insert()" &
    done
    wait

Script inserts 139 M generic text documents ranging from 1 kB to 4 MB according to a probability distribution encoded in the script, with smaller documents being much more common. Average document size is about 60 kB. Documents typically compress about 12:1 with zlib compression in the db.

In three runs script

• hung at ~35%
• reached 50% without hanging when I terminated it
• hung at about 50%

When it hung mongod was consuming 2100% CPU, a majority of that system CPU. Per gdb stacks (attached) I think this is accounted for by 20 threads spinning in __log_slot_join and one thread spinning in __log_slot_switch_internal.