* Implement locks in the file cache
* Replace custom locks
* implement restore lock
Used when re-hydrating
* remove legacy use statements
* Add class descriptions
* Fix style
* Default to database cache driver
* missed cache_locks table
prevent chicken-egg issue
* style fixes
* Remove custom file lock implementation
* missed items from file cache
* Update schema definition
hmm, other schema noise must be from manual modification as this is generated from a freshly migrated db.
* require predis, it is pure python, so no harm in adding
* and set predis as the default client
* Accommodate upcoming php 7.3 change
An announcement on Twitter as well as [this](https://github.com/librenms/librenms/pull/12118) says future release will depend on PHP7.3 minimum. CentOS 8 ships with 7.2 by default. Also, the composer package's availability via `dnf` was dropped somewhere by CentOS project, so one needs to install it "manually" nowadays. I have yet to find a different solution than the one showed here. Any inputs welcome.
* Removed the dnf list command
* Made manual composer install the fallback method
* Fixed minimum PHP version.
* Handle more signals
* Flush buffers before exiting process
This ensures log messages aren't lost
* Restart process before jobs have finished
If there is a very log running job it can cause service restart to
take over 5 minutes.
We tweak the order of things to make sure that running processes
continue, but nothing more is scheduled.
The worst case impact is that a pollling/discovery job gets
scheduled twice, but this should not be a big issue - this should
only occur at most once per day.
* Remove python 3.8 feature
* Ensure that processes from the previous invocation are reaped
* Correct typo's
* Attach subprocess descriptors to /dev/null
Occasionally, PHP would throw a fit and crash when its stdout went
away. To avoid this, we attach stdout to devnull.
This means we lost output of daily.sh - but this is already recorded
in $LOGDIR/daily.log
* Don't immediately schedule long running jobs
To avoid the situation where the maintenance reload happens or a sighup,
then a second long running job is immediately started, we wait
(`last_[poll/discovery]_timetaken` * 1.25) seconds before scheduling
any jobs.
* Add `psutil` to requirements
* Add support for "systemctl reload" to the unit files
* Add a fallback for systems that don't have psutil
* Reduce CPU load when psutil is not installed
* Don't avoid double polling by extending the timeout
This shouldn't happen due to locks
* Remove fallback option
* Remove extra variable
* Fix issue introduced during rebase
* Fix issue introduced when fixing issue introduced during rebase
* Make psutil optional
* Rewrite smokeping script to be an lnms command
* Add a default number of probes for smokeping
* Formatting fixes
* Refactor to simplify a couple of methods
* Fix a name collision when using more than 60 threads
* Simplify
* First pass at documentation
A few climate fixes too.
I'm a little dubious about the + LibreNMS fix - it could be included
in /etc/smokeping/config if it is really needed, but it sounds like
(looking at git blame) it's needed for if the script generates a broken
configuration file.
I'm thinking we should just not generate broken config.
* Improve testability and add some tests
* Load laravel to make translations available
* Second pass at documentation
* Fix brace
* Extend the device factory to include the type
Also adds a device group fake I created before I realised I didn't need it
* Mimic gen_smokeping.php a little more closely
* Update tests to properly verify old and new behaviour against each other
* Replace gen_smokeping with a wrapper
* Don't double whitespace
Render does this automatically.
Explicitly order by hostname too.
* Make faker less likely to generate duplicates
I tried adding a unique constraint here, but it didn't have the
intended effect.
Extending the hostname like this seems to work fine - I tried
generating 10 million hosts and got no duplicates, compared to
duplicates appearing with as few as 100 hosts without this
change.
A true fix would be to add an 'fqdn' fake upstream.
* Make the tests more robust
* Assorted bug fixes
* Style corrections
* Handle 'generic' devices
* Fix an issue that came up during rebase
Co-authored-by: Tony Murray <murraytony@gmail.com>
the entire https://blog.librenms.org site seems to be broken, can't find a new location for this blog post, given the age I assume it has been removed.
* Update Applications.md
Updated docs to resemble updated apache snmp extend script which is now compatible with python3.
* Updated document as requested
* Updated to reflect newest library deps
Changed the agent code to use newer libraries as suggested by SourceDoctor. Docs now reflects the same change.
* Fixed typo
* Small update to trigger stuck tests
* removing $graphs global
* remove unused things
* fix some additional graphs
* Fix graphs persisting too soon
* correct name for poller module performance graph
* only one type of graph is used here
* Initial Release, wireless sensor support for Openwrt
* Extend Openwrt support, add more sensors
* Add documentation for Openwrt Wireless Sensor support
* Moved location of Openwrt.md
* Update openwrt.json
Co-authored-by: Tony Murray <murraytony@gmail.com>