Open chakaz opened 10 months ago
Oh, and continuing (via c
in gdb I mean) the replication process immediately unfreezes everything
yes,this is not surprising, give our we replicate data and wait for Sink::Write() to finish before proceeding in the shard queue.
Perhaps not surprising :) But don't you think we should fix that?
We have only one alternative option - remove the replica if it lags behind too much
BufferedStreamerBase
from io_utils
has clearly defined limits on what gap we can keep before stopping
Perhaps not surprising :) But don't you think we should fix that?
we should, could be something we work in Q1 :)
I ran Dragonfly, attached a replica, and then pushed some data into the master. When doing so, I froze the replica (ctrl-c inside gdb, so the connection is still open). I noticed that the master Dragonfly is partially frozen, in that it does respond to
PING
, but a simpleSET x y
waits forever.More details:
MEMORY STATS
every 1 s, that one is not frozenHere's the output of
DEBUG STACKTRACE
: