Client: Retry a failed config reload instead of failing the running client #117
Reference in New Issue
Block a user
No description provided.
Delete Branch "koolkhel/vitastor:client-reload-retry"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Description
A client reloads the global config when etcd cancels its watch on a compacted revision (
on_reload_hook). Withclient_infinite_startoff, which is the default, one etcd error during that reload goes tofatal_error()and exits the process (or aborts the client whenexit_on_erroris off), although the client is already serving I/O with a valid configuration. Seen with an iSCSI gateway built on the client while etcd was slow (one member down, 600 ms latency on another).The fix: a reload only happens after start, so the reload hook switches the client to
infinite_startand the reload is retried by timer, asload_global_config()already does in that mode.Tested: builds;
test_cluster_clientpasses.Contributor Name
Yury Luneff (MIND Software LLC, https://mindsw.io)
CLA