WL#2669: Slave Cluster recovery should be automated
Affects: Server-7.1 — Status: Un-Assigned — Priority: Medium
If either the slave cluster is being started from a restore of a master cluster dump, or being recoverd from an outage, the slave should have a way of knowing exactly where to start processing from in the master bin log without a user having to complete 4 - 5 step process. (i.e. should have a --recover flag that automates these steps. Conversation from IRC: jeb_phone tomas: and why can't this all be automated? tomas_break well the whole point of it is that it is in the same transaction as the epoch... so it will tell you exactly which epoch was the last successfull tomas_break we have problems convincing brian that epochs are general enough and that is the way to go... tomas_break the whole replicaition is built around binlog positions... jeb_phone so we use the manually instead? Not sure that make sense. ;-) tomas_break but it would be "trivial" to automate... tomas_break almost... tomas_break the binlog_index and apply status are cluster addons... that does not interfere with replication of innodb or myisam... tomas_break if we were to integrate this functionality more tightly it would affect the other storage engines... tomas_break jeb_phone, ^ jeb_phone Just seems that we should be able to tell the slave process to get the last epoch and fing that position in the bin log. That way if we do a restore or have a crash, the slave would know exactly where to take off from without depending on users. <-tomas jeb_phone the more you can remove user intervention the better a product you have :-) tomas_break jeb_phone, but I would have to extend the syntax somehow... tomas_break right now the whole interface is manual... no automation... tomas_break what do do if there is a conflict between apply_status and positions in the master info? tomas_break override? What is the user _want's_ to go with the position in the master info? tomas_break s/is/if jeb tomas: my guess would be to back down the epoch until there is a match in the master. tomas_break I could have a flag update_log_postion_automatically_according_to_apply_status ... tomas_break that would be checked on SLAVE START and correct the log position if there is a discrepancy... tomas_break I only have one issue that does not make it simple... I don't have access to the binlog_index on the master side from the slave... tomas_break jeb, ^ jeb sounds good to me, can you have that before you leave today? Why don't we replicate the binlog_index? jeb tomas ^ tomas jeb, it conflicts with the binlog_index on the slave... jeb Then could we have a master_binlog_index that gets the replicated items pumped into it? jeb Might should get Mats into this conversation. tomas jeb, yes but we are making this more complex than it should be... the replication _should_ support epochs directly... not as an addon... and we enough work to fix what we have already implemented... tomas we need to give people a readon to upgrade to 5.2 :) tomas s/readon/reason
Copyright (c) 2000, 2016, Oracle Corporation and/or its affiliates. All rights reserved.