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
In noncluster environments in a high-availability configuration, there
is only one target server to restore repository data onto. However, the
current implementation performs one rsync task per node in the
replication network. In a setup with one primary instance and two
passive replicas, this would amount to three rsync tasks with identical
or almost identical file lists.
Aside from the rsync task for transferring repository data from the
backup snapshot onto the target server, the other rsync operations per
replica are unnecessary.
Avoiding these redundant rsync tasks reduces the runtime of
ghe-restore-repositories by about 12 % for a customer with roughly 1 TB
of repository data.
With GHE_PARALLEL_ENABLED=yes configured, this change also prevents a
race condition, which could occur when multiple rsync processes try to
write to the same file on the target server at the same time.
0 commit comments