filterx: cache_json_file() inotify based automatic reload #517
Conversation
bshifter
force-pushed
the
fx-cache-file-json
branch
8 times, most recently
from
ea1fd85 to
6656632
Compare
lib/mainloop.c
Outdated
| @@ -167,7 +167,7 @@ struct _MainLoop | |||
|
|
|||
| MainLoopOptions *options; | |||
| ControlServer *control_server; | |||
| CfgMonitor *cfg_monitor; | |||
| FileMonitor *file_monitor; | |||
There was a problem hiding this comment.
Can we keep the instance name cfg_monitor in this unit?
lib/file-monitor.h
Outdated
| void cfg_monitor_start(CfgMonitor *self); | ||
| void cfg_monitor_stop(CfgMonitor *self); | ||
| void file_monitor_start(FileMonitor *self); | ||
| void file_monitor_start_silent(FileMonitor *self); |
There was a problem hiding this comment.
It does not look like a "silent" version of start, it's more like whether you want an "immediate" check or not.
What do you think of start_immediate() or start_and_check()?
Don't you need a bool return value for the event callback? In that case start_and_check() could return that as well.
| if (!self->frozen_objects) | ||
| return; | ||
| self->frozen_objects_history[self->history_index] = self->frozen_objects; | ||
| self->history_index++; |
There was a problem hiding this comment.
Can we add an assert not to overflow the buffer in case the cleanup mechanism fails or someone refactors the guarantees away?
| main_loop_assert_main_thread(); | ||
| if (self->history_index >= FROZEN_OBJECTS_HISTORY_SIZE) | ||
| { | ||
| main_loop_reload_config(main_loop_get_instance()); |
There was a problem hiding this comment.
Can we move out this condition and the reload call to _file_monitor_callback (the name _load_json_file_version does not suggest that such intrusive side-effect might happen)?
Once it is done, can we add a comment to _file_monitor_callback() that this function may reload the configuration, so nothing complicated should be done with config-related references on the caller side?
|
Nice work! :) |
- Support configurable filename - Add support for DELETE event - Renamed to file-monitor Signed-off-by: shifter <[email protected]>
bshifter
force-pushed
the
fx-cache-file-json
branch
from
6656632 to
0c23075
Compare
Due to multiple workers, the original frozen objects must be retained until all workers have finished. Since there is currently no way to track the workers' lifecycle, we call reload, which unfreezes and releases all previously stored elements to prevent high memory usage. This is a suboptimal solution and likely temporary. The upcoming RCU feature will provide a better way to handle this issue. However, the soon-to-be-released native dictionary implementation will also support the optimal solution. Signed-off-by: shifter <[email protected]>
bshifter
force-pushed
the
fx-cache-file-json
branch
from
0c23075 to
671261b
Compare
Due to multiple workers, the original frozen objects must be retained until all workers have finished. Since there is currently no way to track the workers' lifecycle, we call reload, which unfreezes and releases all previously stored elements to prevent high memory usage. This is a suboptimal solution and likely temporary. The upcoming RCU feature will provide a better way to handle this issue. However, the soon-to-be-released native dictionary implementation will also support the optimal solution.