Age | Commit message (Collapse) | Author |
|
partitions
Partition is never created if it is already known to be obsolete (and so to be dropped).
|
|
|
|
|
|
|
|
|
|
partitions" bug
but really :-)
|
|
Truncate events can be ignored on per handler basis.
|
|
|
|
|
|
partitions" bug
|
|
partitions" bug
|
|
|
|
develop
|
|
|
|
|
|
|
|
|
|
(in response to "londiste3 repair command floods the console with the diff mesages")
|
|
|
|
|
|
|
|
|
|
This gives more information and is less random.
|
|
|
|
|
|
|
|
It was getting more and more confusing, especially since 'dispatch' handler mixed both partitioning types (table and database) in one class.
|
|
(dropped) partitions
|
|
|
|
|
|
|
|
|
|
|
|
|
|
The problem appears in 'catching-up' phase were code checked
table status only and not whether the table is owned by
current process. Then events can be replayed by several copy
processes when they are all in catching-up phase.
The problem does not appear when local_only=1.
Also remove unused function.
|
|
|
|
|
|
|
|
Way to add extra connect string parameters for non-config
connection strings (public node locations).
|
|
|
|
otherwise they conflict with change operations.
|
|
|
|
|
|
|
|
|
|
|
|
Note that current behaviour is that dispatcher would crash if events older than retention period arrived.
Perhaps it is better than quietly recreating and redropping tables for too old events..?
|
|
dropped tables
|
|
It controls how long to keep partitions around. Examples: '3 months', '1 year'
|
|
|