This happens with some frequency when testing with higher numbers of threads/batchSize.

e.g. processingThreads=40, ingestionThreads=2, batchSize=6144

fails with "restore error: error writing bulk insert: write tcp 10.4.125.156:27017: broken pipe"

after 3638.31824303 seconds (over an hour).

For comparison a successful run completes in under 10 minutes.

mongorestore log: https://gist.github.com/amidvidy/1ea1deee73bf82885af3
mongod log: https://gist.github.com/amidvidy/8c25c445770f7272f5cb

I have bunch more logs of identical failures if those are helpful.