Default value for replication_sync_timeout
Having a non-zero replication_sync_timeout gives a user the false assumption that the box.cfg{replication = ...} call returns only when the configured node is synced with all the other nodes.
This is mostly true for the big replication_sync_timeout values, but it is not 100% guaranteed.
In other words, a user still has to check if the node is synced, or the sync just timed out.
Besides, while replication_sync_timeout is ticking, you cannot reconfigure box with another box.cfg call, which hardens reconfiguration.
It is decided to set the replication_sync_timeout to zero by default.
The compat module allows you to choose between
- the old behavior:
box.cfg.replication_sync_timeoutis 300 seconds by default - and the new behavior:
box.cfg.replication_sync_timeoutis 0 by default.
It is important to set the desired behavior before the initial box.cfg{} call to take effect for it.
tarantool> compat.box_cfg_replication_sync_timeout = 'new'
---
...
tarantool> box.cfg{}
---
...
tarantool> box.cfg.replication_sync_timeout
---
- 0
...
tarantool> compat.box_cfg_replication_sync_timeout = 'old'
---
- error: 'builtin/box/load_cfg.lua:253: The compat option ''box_cfg_replication_sync_timeout''
takes effect only before the initial box.cfg() call'
...
A fresh Tarantool run:
tarantool> compat.box_cfg_replication_sync_timeout = 'old'
---
...
tarantool> box.cfg{}
---
...
tarantool> box.cfg.replication_sync_timeout
---
- 300
...
We expect issues with a user assuming that the node is not in the orphan state (box.info.status ~= "orphan") after the box.cfg{replication=...} call returns.
This is not true with the new behaviour. To simulate the old behavior, one may add a box.ctl.wait_rw() call after the box.cfg{} call.
box.ctl.wait_rw() returns only when the node becomes writable, and hence is not an orphan.