SEC.CFG.PATCH_EOL — The server runs a release series that is out of support
- Category: security
- Severity: high
- Level: 0
- Confidence: deterministic
- Downtime class: none — the finding is about the server's version, not about a statement
- Stability: stable
- Suites: audit
- Applies to: PostgreSQL 18, MySQL 8.4
The finding is about the future, which is why it does not wait for a CVE
An out-of-support server is not insecure the moment support ends. Nothing changes on that day: the same queries return the same rows, and no scanner finds anything new.
What ended is the supply of fixes. When the next vulnerability in that engine is found — and one will be — supported series receive a patch and this one does not. The interval between a fix existing for everybody else and an exploit existing for the rest is measured in days.
So the rule reports a countdown that has already started, rather than a defect you can reproduce. That is also why there is no configuration change that answers it: the remediation is a major upgrade, planned and scheduled.
Why high and not critical
critical is reserved for exposure that has been demonstrated — a credential in the open, a grant
that hands the database to anybody who asks. Nothing here is known to be exploitable today.
A team that has to plan a major upgrade needs to know it is overdue. Telling them the building is on fire spends the weight that a real fire will need.
Ahead of the end, a quieter finding
Within sqlens.security.advisories.lead_days days of the end, 180 unless you set another number,
the rule reports the series as ending: at medium, with the date and the days left. Nothing is
wrong with the server yet, and it receives fixes until then. But an upgrade across a major is a
project rather than a patch, and a finding on the day after support ends comes too late to plan
one.
The end date itself still counts as supported and belongs to the lead. From the next day on, the
finding is the high one above. 0 turns the lead off, so the rule reports a series only once
its support has ended. The shipped security presets block on high and above, so the lead does
not stop a build; it shows up beside the findings that do.
What the rule reads
-- What the server answers, and what the rule reads.
SHOW server_version; -- 13.22
-- PostgreSQL 13 left support on 2025-11-13: no further minor release
-- will ever carry a fix for it.
-- The same question on a series still inside its window.
SHOW server_version; -- 18.4
-- PostgreSQL 18 is supported until 2030-11-14.
There is no SQL that fixes this, which is unusual for a finding and worth saying plainly: the remediation is a major upgrade, planned and scheduled, and no setting substitutes for it. The examples show the READING rather than a repair because that is the entire surface a database exposes here -- the rule judges what the instance says it is.
Where the dates come from, and what they can and cannot say
The verdict is reached offline, from resources/data/eol.json, which this package ships. That is
deliberate: a check that had to reach the network would be unusable on a CI runner without egress,
and would answer the same question differently on two days.
Three files can answer, in a fixed order:
- a path you configure in
sqlens.security.advisories.path resources/data/eol.jsonpublished into your own application- the copy bundled with the package
The file that actually answered is named in the finding. Without that, a refresh that landed somewhere nothing reads would look exactly like a refresh that changed nothing.
The dates record premier support for MySQL rather than extended. Extended support exists, is purchased separately, and is not what most projects are on — recording the later date would let a server that stopped receiving routine fixes read as supported.
What it says when it cannot tell
Three answers are not findings about your server but statements about the check:
| Situation | What you get |
|---|---|
| the series is not in the data at all | undetermined — the server is newer than the file, or older than anything it records, and neither is a verdict |
| the version string could not be read | undetermined with setting_value_unrecognized, naming the string. A pin changes nothing here, because this check judges the live value rather than the version a run reasons about |
| no advisory file could be read | undetermined, naming which path was tried and why it failed |
None of them is a pass. A version with nothing to hold it against is not a version in support.
False positives
A vendor-patched build that carries security fixes past the upstream date reads as out of support here, because the version string is all a server offers and it does not distinguish a distribution's backports from upstream's own releases. That is the honest limit of an offline check: the alternative would be trusting a build suffix, which anybody can set. Such a server is suppressed through the baseline once, with the reason recorded.
Related
- SEC.CFG.TLS_DISABLED — another finding about the server itself rather than about anything a migration wrote.