Comment 2 for bug 1455810

Taking the backups (those commands stream to a remote location):

Full:
innobackupex --defaults-file=XXX --compress --user=XXX --throttle=50000 --parallel=1 --stream=xbstream --tmpdir=XXX --no-lock work_dir

Differential:
innobackupex --defaults-file=XXX --incremental --incremental_lsn=XXXX --compress --user=XXX --throttle=50000 --parallel=1 --stream=xbstream --tmpdir=XXX --no-lock work_dir

The defaults file is our regular my.cnf file plus backup user credentials.

Starting with expanded decompressed backups in work_dir (full) and differential_dir:

innobackupex --apply-log --redo-only --use-memory=32G work_dir 2> full.log
innobackupex --apply-log --incremental-dir incremental_dir --use-memory=32G work_dir 2> differential.log

The first command works fine, the second one is the one raising signal 11.

Full backup size is 1.36T (compressed, so ~2.2T after expansion).
Differential is 76G compressed.

Also, we do hundreds of restores a week of various databases using that exact process. We encountered that issue twice on two distinct differentials of the same source DB (and sharing the same from_lsn/full).