Stats are pushed to https://solomon-test.yandex-team.ru/?project=junk&cluster=foo&service=bar&l.context=migration_*

* Stop coremons/frontends on target cluster
* Check and fix config.sh (Note: it is deployed during build_*.sh, so edit it now)
* Optionally: drop all metabase tables on target cluster. wint@ has scripts to quickly format any target cluster.

* Run build_mb.sh. This will build and deploy all necessary files to source hosts
* Edit start_mb.sh, set PERIOD variable to zero (to transfer all the metabase)
* XX: Run start_mb.sh
* XX: 'watch -n 15 check_mb.sh' - this will produce some stats on the number of finalized shards (and number of sensors in them) and will also push stats to solomon-test
* XX: Wait till number of processes runs down to zero - this takes up to 12 hours. If something breaks up, comment out last line in start_mb.sh and run it to kill all the migration processes.
* Check that total number of sensors migrated is equal the total number of /FileSensors on source cluster. Check that no logs on remote clusters contain stacktraces.
* Run the XX: steps with PERIOD set to 1: this will migrate all the sensors created in the last 1 day (i.e. while it migrated all the older sensors)

// Metabase is now migrated!

* Run build_sp.sh. This will build and deploy all necessary files to target hosts
* ssh to any target host, and run '~/migrate/run_sp.sh --mode MAKE_STAMPS all' (note: there are some other useful modes available, check StockpileShardTransfer)
* Restart stockpile on all target hosts - all stockpile shards should now pause in INIT state.
* Start coremons on target clusters (and fetchers, if you stopped them for some reason). Once they fully load - please check, that number of /FileSensors on target cluster is correct (~ numer on the old cluster).
* Run start_sp.sh (if something breaks up, comment out last line in start_sp.sh and run it to kill all the migration processes)
* Wait until all stockpile shards on target cluster are DONE - this takes up to 28 hours. If graphic stops rising earlier - something went wrong, check logs.

// Stockpile is now migrated!

* Start frontends on target cluster and check that everything is really fine.
