App::Wubot::Guide::Debugging - guide for debugging feeds
The following document attempts to describe some steps you can take to create a new reaction or debug a reaction that isn't working properly.
The wubot-check script can be used to run a single check of an instance of a monitor. Simply pass in the name of the plugin and the name of the instance config file, e.g.:
wubot-check RSS usgs-m5
This would attempt to read a file:
It will actually search for any file in the RSS directory that begins with usgs-m5.yaml, so even if you have disabled the plugin instance by renaming the .yaml file to .yaml.disabled, it will still be loaded.
The check instance's cache file will be read and the check will be executed once, and any reactor messages generated will be dumped to stdout. So if you have previously run this instance of the monitor, it's cache data may prevent it from re-sending any messages. See the 'cache file' section below.
Any reactions defined directly in the monitor config will be executed, so this can be a great way to debug simple reactions like setting or modifying fields on the message.
Many monitor plugins cache data that has been seen in a cache file. The cache files live here:
For example, the cache file for RSS-usgs-m5 would live here:
Once this cache file has been created, if you run the monitor again, it may send through no data. To re-send feed data through for debugging, it may help to remove or temporarily rename the cache file.
If deleting the cache file sends too much data to see what is happening, then you could edit the cache file, remove a single entry, and then re-run the wubot-check script to trigger that item to get re-processed.
When the trouble is occurring in the reactor config, sometimes seeing the raw monitor data using wubot-check isn't so helpful. In such cases you can use the Dumper reactor plugin to dump out the entire message contents. At the point of confusion, simply insert a rule like this:
- name: dump message contents plugin: Dumper
Note that where you put the rule may make all the difference. For example, if you have a rule that modifies the message, then using the Dumper rule before that rule would show you the contents before the message was modified, and using the Dumper rule after the message modifying rule would show you the modified content.
Both wubot-monitor and wubot-reactor write logs to ~/logs. The logs are rotated a midnight each night. Only the last 10 days of logs are stored. Both processes default to the 'info' level logging on stdout and 'debug' level logging to the log files.
To enable debug logging to stdout, export the environment variable LOG_DEBUG. For even more debugging info, export LOG_TRACE. For example:
LOG_DEBUG=1 bin/wubot-monitor LOG_TRACE=1 bin/wubot-reactor
Message queues in wubot are implemented using a SQLite database. The contents of the message is serialized using YAML::XS and stored in the SQLite db.
After a message has been processed successfully, the row will not immediately be deleted. Instead, wubot will update the row to set the 'seen' column to the time the message was processed. So you generally have at least a 24 hour window where you can go back an inspect the message data that was sent to the reactor in the event of a problem. You can even re-send the message through the queue by hunting down the id and setting 'seen' back to null. This will cause the reactor to immediately pick the message back up from the queue. For example:
# open the default reactor queue sqlite3 ~/wubot/reactor/queue.sqlite # find the most recent message from the USGS rss feed SELECT * FROM message_queue WHERE data LIKE '%RSS-usgs-m5%' ORDER BY lastupdate DESC LIMIT 1; # that gets the one we want, so get the id SELECT id FROM message_queue WHERE data LIKE '%RSS-usgs-m5%' ORDER BY lastupdate DESC LIMIT 1; # now set 'seen' back to null UPDATE message_queue SET seen = NULL WHERE id = 21991;
Alternately, if you have a plugin that has sent through a flood of messages (or you temporarily created a routing loop between bots), the reactor may take a very long time to crank through all the messages in the queue. So you could go in and mark messages as 'seen' or even delete them if there are a huge number.
# delete the messages from the queue - be careful here update message_queue set seen = 1 where data like '%Some Flooded Message Text%'; # delete the messages from the queue - be careful here delete from message_queue where data like '%Some Flooded Message Text%';
Each time wubot opens a new connection to the database, it will purge all 'seen' messages that were marked seen more than 24 hours ago. This ensures the message queue will not get too long, as very long message queues will significantly degrade performance. The performance seems to start to degrade at around 50k messages, and may use significantly more system resources if the queue grows beyond 100k messages. Your mileage may vary. After purging old seen messages, it will log the number of seen and unseen messages left in the queue.