When searching large log files for messages that do not frequently occur, there is a large risk of reading all collected lines before returning the first results.
In most cases recent items are required, in which case going back for days of logs might not make sense.
This commit adds a simple "history" selection which translates into a "valid_from" filter on the log data. When timestamps are not parseable for whatever reason, the filter is ignored.
Only small downside is that we do need to translate timestamp again, when needed we could improve performance a bit by storing the original datetime value in NewBaseLogFormat so we don't have to parse it twice.
It only matters for lighttpd preventing a bug but since this could
be copied and some service being added has the same issue make sure
the code is always the one that's least problematic which means
putting "lo0" as the first explicit (and automatic) interface.
One of the nice things is that we don't have to do the global
loop but get all of the information across at once even if
individual listeners will still loop over it if necessary.
Dashboard is now stored for each user so to synchronise the user
sync needs to be used.
While here remove the old structure from the default config.xml.
PR: https://forum.opnsense.org/index.php?topic=42392.0
Since the default argument count is ":1" these listeners never
see interfaces so they can be called over the new event and the
old one can be deprecated.
Some services are global so calling them to reconfigure for each interface
is suboptimal. We can move these to a mapped event with all the interfaces
to reload at once which avoids useless cycles and race conditions.
Tried to check PID and avoid start without family association
but extending to pass the address family made more sense so we
can always restart to ensure addresses and routes are matching.
More work needed but as discussed with @swhite2 we want this here
before we work on the UX part. These types of notifications need
more screen presence and are non-dismissable.
While here kill an oddity assignment with constant() use.
And adjust comments still refering to "LiveCD" since we have a DVD
image nowadays and the live mode works on all install media.
This probably needs more engineering later on which makes
it a little harder to ship but in reality we want people
to run the portable command easily in order to unbreak their
environments as pkg-upgrade tripping over its own temporary
files and directories keeps happening in user reports.
While here use "-s" while at it to speed up the script call.
This also pertains to "-e" which is its sibling.
If widgets fail to load for any reason, the tick routine simply continues
without interruption. This commit makes sure that the visual error
state isn't stuck and switches back to widget output when possible.
The easiest way to reproduce this is to simply disconnect from
the network while having the dashboard open.
Minor temporary fix in case /var/db/rrd/updaterrd.sh doesn't exist. All of this will vanish soon, so it isn't worth the effort of chaging the file mode and retesting.
ref https://github.com/opnsense/core/issues/7753
