Alert when your machine is becoming oversaturated, *before* it slows to a crawl
tl;dr: psi-notify can alert you when resources on your machine are becoming oversaturated, and allow you to take action before your system slows to a crawl.
psi-notify is a minimal unprivileged notifier for system-wide resource pressure using PSI. This can help you to identify misbehaving applications on your machine before they start to severely impact system responsiveness, in a way which
MemAvailable, CPU graphs, I/O utilisation graphs and other metrics cannot.
CONFIG_PSI(enabled by default in most distributions)
On Arch, the psi-notify AUR package is available.
On Fedora and RHEL/CentOS 8, the psi-notify package is available in Fedora/EPEL.
Otherwise, manual installation is as simple as running
makeand putting the resulting
psi-notifybinary in your PATH. You will need
After that, you just start psi-notify. A systemd user service is packaged and can be used like so:
systemctl --user start psi-notify
Put your configuration in
~/.config/psi-notify. Here's an example that will check roughly every 5 seconds⁺, and pop up a notification when the values are exceeded:
update 5 log_pressures false
threshold cpu some avg10 50.00 threshold memory some avg10 10.00 threshold io full avg10 15.00
The above is the default configuration if no config file exists. You may have to tweak these depending on your hardware, machine behaviour, and normal workloads.
You can reload the config without restarting by sending
Look at the "config format" section below to find out more about what a valid config looks like.
⁺ PSI has poll() support, but it's not currently available to unprivileged users. See this discussion.
oomd and psi-notify are two compatible and complementary projects -- they're not in opposition to each other. oomd also uses PSI metrics, but it requires a policy about "what to kill" in high-pressure scenarios. For example, on a web server we obviously don't want to kill the web server if we can avoid that, so we should prioritise other applications. On the desktop though, it's hard to say: should we kill Chrome, or some IDE, or maybe something playing a movie? It's extremely difficult (although perhaps possible) to produce a single configuration that will do the right thing in even the majority of cases, so we opt to alert early instead and have the user make the decision about what's high priority in their user session. When integrating oomd for the desktop, most distributions will likely end up having to make it less aggressive than would be ideal, so they can still interoperate.
It's hard to produce a good policy for, say, one's working day, where at one time my terminal is the most critical thing, at another my browser is, and at another it's my mail client. At other times maybe I'm ok with the slowdown and am willing to ride it out without killing anything. psi-notify sidesteps this problem by simply notifying, rather than taking action.
The update interval in seconds is specified with
update [int]. The default is
update 5if unspecified.
If you'd like messages like this at every update interval, you can set
log_pressures true(the default is
INFO: Current CPU pressures: some avg10=0.00 avg60=0.02 avg300=0.01 INFO: Current memory pressures: some avg10=0.00 avg60=0.00 avg300=0.00 INFO: Current memory pressures: full avg10=0.00 avg60=0.00 avg300=0.00 INFO: Current I/O pressures: some avg10=0.00 avg60=0.00 avg300=0.00 INFO: Current I/O pressures: full avg10=0.00 avg60=0.00 avg300=0.00
Thresholds are specified with fields in the following format:
ioare currently supported.
fullmetric. See the definition here.
avg300are currently supported.
Issues and pull requests are welcome! Please feel free to file them on GitHub.