The function udev_enumarte_scan_devices returns a value less than 0 on
failure. If this is the case then we terminate the read for this smart
information.
This change was already send upstream. And could be delete in feature
collectd versions.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
This was not a real fix but a workaround. It is no longer clear to me
why this was necessary. Deleting the patch restores the upstream
behaviour of the collected for the smart plugin. I have tested it and on
my system the hard disk to be monitored is recognised.
root@system ~ # cat /sys/class/block/sda/uevent
MAJOR=8
MINOR=0
DEVNAME=sda
DEVTYPE=disk
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
On my system the attribute DEVTYPE was not set. The plugin could not
read any data and the function call blocked forever on this function and did
not returned. By removing it, all block devices under `/sys/class/block`
were checked.
Block devices that do not support SMART were not evaluated. The
collected displays the following message.
smart plugin: checking SMART status of /dev/loop4.
smart plugin: unable to open /dev/loop4.
If you do not like this message, you could only enable device in the uci that
does support SMART.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
The crude loop I wrote to come up with this changeset:
find -L package/feeds/packages/ -name patches | \
sed 's/patches$/refresh/' | sort | xargs make
Signed-off-by: Ilya Lipnitskiy <ilya.lipnitskiy@gmail.com>
Add "ipstatistics"-plugin. This plugin parses "/proc/net/netstat" and
"/proc/net/snmp6" to get the overall ipv4 and ipv6 usage.
Signed-off-by: Nick Hainke <vincent@systemli.org>
Getifaddrs() is not working all the time (e.g. wireguard interfaces).
Instead we use if_nameindex() syscall as suggested in:
https://stackoverflow.com/a/45796495/8474618
Signed-off-by: Nick Hainke <vincent@systemli.org>
Count the current dhcpleases. Currently, we use a bash script
that does the same job (Freifunk Berlin). We want to use native
collectd plugin for that.
Signed-off-by: Nick Hainke <vincent@systemli.org>
Update collectd to version 5.12.0.
* Refresh patches
* Remove upstreamed patch 920-fix-ubi-data-source-type.patch
* Backport a patch from upstream to fix a missing define
(920-backport-netlink-reg-noerror.patch)
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Currently the values are stored in the ubi plugin as data source type
`counter`. But this makes no sense, because the values change very slowly
and I don't want to know the rate of change. It is better to store the
value as data source type `gauge`. Then I can see the current value.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
Add a script that can collect data from mq (multi queue) and cake
qdiscs.
Script is reliant on collectd's 'exec' module.
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
Update collectd to 5.10.0
* leave new plugins as disabled for now (procevent, sysevent)
* refresh patches. Remove unenecessary version fix
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Adjust the reaction to a polling interval timestamp that references
to a past time.
Past timestamps can happen when ntpd adjusts router's time after network
connectivity is obtained after boot. Collectd shows warnings for each plugin
as it tries to enter new values with the same timestamp as the previous one.
This patch adjusts the next polling time to be now+2 seconds for the main
loop and for the plugin-specific read loops. That avoids the warnings, but
does not overreact in case there are shorter polling intervals or the time
gets adjusted for other reasons.
Additionally some debug statements are aded, but they are visible only
when --enable-debug configure option is used in Makefile.
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Upstream project's behaviour of reporting this data source in range 0-1
is used by at least CentOS 7 collectd 5.8.1, very likely the same
behaviour is also in widely use by other distributions.
The change is important for providing consistent experience across
collectd deployments on different environments
Ref: https://github.com/collectd/collectd/pull/3244
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
Update collectd to version 5.9.0
* remove backported upstream patches (sensors, modbus)
* remove lua patch that has been been implemented upstream
(See collectd PR 3144)
* refresh patches and accommodate into upstream path changes
* place the 'stations' datatype for iwinfo in alpbetically
correct place in types.db
* add libmnl dependency to 'processes' plugin required for the
new Linux process Delay Accounting capability.
(Alternatively the capability & requirement might be patched away
in Makefile.am, HAVE_LIBTASKSTATS from collectd_4ea7a572)
* new plugins (disabled): ampq1, gpu_nvidia, pcie_errors,
write_stackdriver, write_syslog
compile-tested: ar71xx/WNDR3700 (all plugins)
run-tested: ar71xx/WNDR3700 (selected plugins)
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
If you want to register more than one lua plugin, you cannot start the
collectd.
You get always the error message:
"The read function "lua/(null)" is already registered. Check for
duplicates in your configuration!"
This is not what I expect and what the documenation says.
With this change it is now possible to registrate more then one lua
Script.
See https://github.com/collectd/collectd/pull/2379
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
* 938897a2 Add scale and shift to modbus plugin
* 60280b80 correcting all the wrongs
* a00ab529 Add support for RS485 to modbus plugin
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Backport an upstream patch to remove the upper limit of
lm-sensors version check, so that using lm-sensors 3.5.0
will work.
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Support for little-endian word-order of 32-bit fields was added
upstream recently. Import the corresponding PR
https://github.com/collectd/collectd/pull/2660
into a local patch and bump package release.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
This patch upgrades the collectd package from 5.7.2 to 5.8.2.
All openwrt patches got either updated or deleted in case they
are already included into upstream collectd.
The new collectd 5.8.0 package compiles and got tested on current
lede trunk with latest luci on an PCengines APU. All default plugins
(enabled via Openwrt Makefile) - so also iwinfo - are running fine
with the new version.
Signed-off-by: Thomas Huehn <thomas@net.t-labs.tu-berlin.de>
Amended with the following changes:
* preserve git file history by avoiding unnecessary patch renames
* mark new plugins introduced with 5.8 branch as disabled
Compile tested all plugins with ipq806x
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Uptime plugin fails to adjust for system time changes after boot.
As Openwrt/LEDE routers usually do not have a RTC, the system time
gets adjusted with NTP possibly after collectd has already started.
But collectd continues to use the initial time set by 'sysfixtime',
which can lead to incorrect uptime calculations.
Apply a proposed fix from upstream that uses /proc/uptime
Reference to https://github.com/collectd/collectd/pull/2034Fixes#4819
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Buildbot revealed that Collectd 5.7.1 uses libcap to evaluate system
capabilities and display advice on some plugins (dns, iptables, exec,
ping, ceph, turbostat).
Disable the check in configure.ac in order to avoid a dependency
for libcap. (It is not that big, only 6 kB .ipk, but still avoid that.)
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Update collectd to version 5.7.1
* replace 003-remove-werror.patch with configure option
* remove 200-fix-git-describe-error.patch that is now unnecessary
* reformat other patches due to whitespace changes
* add new plugins to the list as disabled
* correct download & homepage links
* add myself as the second maintainer
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Update collectd, the backbone of Luci statistics, to 5.5.1
Refresh patches. Main changes:
- Remove 500-upstream-parallel-build-fix.patch (implemented upstream)
- Add 600-fix-libmodbus-detection.patch to revert an upstream change
that broke libmodbus detection in collectd's configure script
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Some collectd plugins launch third-party plugins that have variation in
initialisation time (like libpcap). Recently (since kernel bump to 4.1)
the DNS plugin has been causing collectd to crash semi-randomly at startup
on MIPS based WNDR3700.
Debugging led to realisation that the DNS plugin seems to require at least
0.1s time to start, before the first data reading attempt starts.
By default, the first data read cycle starts immediately, while apaprently
some of the plugins may still be asyncronously initialising. To make things
safe, this patch adds 1 second delay before the first data read cycle.
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Update collectd, base of Luci statistics, to 5.5.0.
Patches have been refreshed.
write-graphite plugin is now enabled. (see #1351)
I have compile-tested all plugins with ar71xx,
and real-life tested the following plugins:
Conntrack, Processor (CPU), DNS, Interfaces, Wireless, System Load,
Memory, Ping, Uptime
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Update collectd, base of Luci statistics, to 5.4.2.
Patches have been refreshed.
905-fix-sigrok-upstream-patch-post-541 was removed as unnecessary.
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
-update collectd to the current release 5.4.1
-fix sigrok glib check (upstream patch after 5.4.1)
Note:
Changes are required also to luci-statistics in Luci source.
Patch in http://patchwork.openwrt.org/patch/5303/
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Alex Shumsky