When a service dies on an EL10 box, the reason is usually already on the
host. It's just in three places. The journal has the failure, the audit log
has the AVC denials, and /proc/pressure knows the box was short on memory
at the time. Sentry-v2 reads all three as they happen, ties each unit
failure to the denials that actually belong to it, and writes the lot up as
one redacted JSON bundle.
If you let it, it'll also throttle or freeze the job that's crowding everyone else out, and put it back when the time's up. It won't do that until you've switched it on twice: once for the daemon, once for the policy.
That's examples/demo/run.sh
on Rocky Linux 10 with SELinux enforcing. Two services are denied the same
file in the same second, and one of them dies. Sentry-v2 links the dead one
to its own denial at 0.976, because the PID was in the unit's cgroup, the
domain is the unit's and the binary matches ExecStart. The other denial
stays out of the bundle. Here's
the bundle it wrote.
- Failures linked to their denials, scored, with the reasons spelled out. A denial from a process in another unit's cgroup isn't counted, even when it's the same binary in the same domain.
- Pressure from PSI triggers, so the kernel wakes Sentry-v2 the moment a cgroup crosses the stall threshold you set.
- Actions you can take back: a runtime
MemoryHighandCPUWeightthrottle, freeze and thaw, or a backup of the unit's files and a reload. Every action is written to disk before systemd sees it. Throttles and freezes expire on their own, and roll back on stop and after a crash. - Bundles where you already look: on disk, an HMAC-signed webhook, RFC 5424 syslog, Splunk HEC or Elasticsearch.
- Packaged for EL10: an RPM with a hardened unit, its own enforcing
SELinux domain,
sentry-v2ctl, a Cockpit page and an optional Prometheus endpoint. Nothing at runtime comes from outside BaseOS and AppStream.
Rocky Linux, AlmaLinux, RHEL or CentOS Stream 10. Grab the RPMs from the latest release, check them, and install:
$ sha256sum -c SHA256SUMS --ignore-missing
$ gh attestation verify sentry-v2-0.2.10-1.el10.noarch.rpm --repo jamesconwaydev/sentry-v2
# dnf install ./sentry-v2-0.2.10-*.noarch.rpm ./sentry-v2-selinux-0.2.10-*.noarch.rpmEL kernels ship with PSI off. This turns it on, and only reboots a host that doesn't have it yet:
# test -e /proc/pressure/memory || { grubby --update-kernel=ALL --args=psi=1 && reboot; }Only want to check policies in CI? pip install sentry-v2 gets you
sentry-v2ctl policy validate without the daemon.
The RPM drops three example policies in /etc/sentry-v2/policies.d, all in
dry-run. Copy one, check it, start the daemon:
# cd /etc/sentry-v2/policies.d
# cp 20-unit-failure-notify.toml 50-myapp.toml
# sed -i 's/^name = .*/name = "myapp-notify"/' 50-myapp.toml
# sentry-v2ctl policy validate
ok: 4 policies
# systemctl enable --now sentry-v2
# sentry-v2ctl status
sentry-v2 0.2.10 running for 3s mode dry-run
COLLECTOR HEALTH
psi state=ok, triggers=0, restarts=0, last_error=None
journal state=ok, restarts=0, last_error=None
audit source=netlink, user_avc_truncated=0, overruns=0, rcvbuf=425984, restarts=0, last_error=NoneThen break something on purpose and read what it wrote. The quickstart walks you through it.
It never restarts, stops or kills a service, and it never touches SELinux
policy or booleans. audit2allow is still your call. SSH, journald,
D-Bus, logind, auditd, your login sessions and Sentry-v2 itself are
protected, and so are whole slices unless you say otherwise. A unit that
needs three actions in half an hour is put on hold until a person looks
at it. And in dry-run, where it starts, it changes nothing at all.
The full rules are in Actions and safety, and where the scoring can mislead you is in the FAQ.
Everything else, from every config key to the bundle schema, is in docs/. For recipes that start from a problem on your host, like tying a dying service to its denials or throttling a batch job that starves the box, see the wiki.
Questions and ideas go in Discussions. If you'd like to write some code, the good first issues each name the file, the test and when it's done, and none of them needs root or a test VM. CONTRIBUTING.md has the rest.
Apache-2.0. Maintained by James Conway (@jamesconwaydev); how decisions get made is in GOVERNANCE.md.
