Panic when we are sending an error to our master/slave.

Related to #5135, see discussion there.
This commit is contained in:
antirez 2018-07-17 17:42:25 +02:00
parent e03358c0d9
commit afc7e08a20

View File

@ -349,6 +349,11 @@ void addReplyErrorLength(client *c, const char *s, size_t len) {
serverLog(LL_WARNING,"== CRITICAL == This %s is sending an error " serverLog(LL_WARNING,"== CRITICAL == This %s is sending an error "
"to its %s: '%s' after processing the command " "to its %s: '%s' after processing the command "
"'%s'", from, to, s, cmdname); "'%s'", from, to, s, cmdname);
/* Here we want to panic because when an instance is sending an
* error to another instance in the context of replication, this can
* only create some kind of offset or data desynchronization. Better
* to catch it ASAP and crash instead of continuing. */
serverPanic("Continuing is unsafe: replication protocol violation.");
} }
} }