| Age | Commit message (Collapse) | Author |
|---|
|
... thus we simply read the configuration the first time it really becomes
necessary and a message requireing the data needs it actually.
(cherry picked from commit bcfe967f607a83192d75c01e7f414655891eec60)
|
|
While mangling the config dict retrieved via get_config_dict() into a private
representation of a configuration dictionary sync-groups were never accounted
for. Instead everything always ended up in the regular vrrp transition-script
section.
The implementation has been changed to directly work on the content of
get_config_dict() to stop any confusion and making redundant data copies
obsolete.
(cherry picked from commit 025f0609cea8591e93b8cb4a7d0256e43e23323b)
|
|
Do not create rfc3768-compatibility interfaces by default because of wrong
Jinja2 syntax. Backporting the entire system makes it easier in the future to
additional bugfixes.
|
|
Move keepalived configuration from /etc/keepalived to /run/keepalived.
(cherry picked from commit b243795eba1b36cadd81c3149e833bdf5c5bea70)
|
|
|
|
|
|
In old implementation, all notify scripts are configured directly inside a keepalived.conf. This way is not recommended even by keepalived as scripts execution is not tracked and they may be running not in order and killed before the actual work will be finished. We have observed such situations in very rare cases during tests.
New implementation use FIFO pipe, to which keepalived send any state changes. And these notifications are read by a daemon, saved in a queue and processed one by one, which eliminates the situation when the scripts creating inconsistent config or environment.
|
