we found diskI/O getting very high and a careful check shows "sync" in a container may cause very high disk I/O behavior. even cgroup blkio.throttle cannot supress the I/O.
we also find that, if we do some disk I/O critical task in a container, and meanwhile do "sync" on another, the sync in the latter will be heavily delayed. e.g., in CT1 we do
stress -d 4
while in CT2 we do
time sync
then the sync would run more than 40 second. (for comparison, when we do nothing in CT1, CT2's sync costs only 1 second about).
could anyone confirm if the "sync" call in container would cause the whole host into sync or not?