An IIS error page tells you nothing until you read the sub-status and the pool state. Decode
both before anyone recycles a pool or reinstalls anything.
Climb the Troubleshooting Ladder base skill first: past tickets for this site — a
deployment, a certificate renewal, a Windows or .NET patch, a service-account password
change (a pool whose identity password changed won't start) — then the documented server:
IIS version, whether a vendor app is hosted here, pool-to-site mapping, runtime version,
pool identity, expected bindings and certificates.
Evidence before theory: exact status and sub-status, pool Started/Stopped, the System log
for WAS and rapid-fail entries, the Application log for the app's exception, HTTP.sys logs.
For an intermittent 500, enable Failed Request Tracing. Never act on the browser's generic
"500".
Branch:
a. Pool crashing / rapid-fail — Stopped after WAS disabled it on N failures in the
rapid-fail window: the worker process is dying. Read the Application log or crash dump
for the exception — bad deployment, missing dependency, runtime mismatch, memory. Never
fix a 503 by re-enabling a pool rapid-fail disabled; that hides a crashing app. Vendor
code goes to the vendor with the exception.
b. Bindings and SSL — HTTPS failing, wrong certificate, or handshake errors after a renewal.
A renewed certificate has a new thumbprint and often was never re-bound; SNI-versus-IP
bindings and host/port collisions do the same. Check the HTTP.sys SSL binding, not just
the IIS UI, and treat certificate work as a change (see the SSL certificate renewal
playbook).
c. 500-series decode — 500.19 is configuration/web.config (bad section, locked config, file
permissions); 500.21 means a runtime or module isn't registered; 500.0 and 502.5 mean the
app or runtime process fails to start. Confirm the sub-status against its documented
meaning on the web and fix that layer only.
d. 503 — almost always a stopped or disabled pool (branch a) or HTTP.sys with no listener
for the binding. Confirm pool state first; it is infrastructure, rarely app code.
e. Recycling — constant recycling on interval, memory-limit or config-change triggers drops
sessions and warms slowly. Read the recycle events: intended trigger, or a memory leak?
Tuning a vendor app's recycling follows vendor guidance.
For a vendor application, never edit its web.config, pool runtime or identity without vendor
guidance — you break support and the app. Hand config changes to the vendor or app owner.
Restarting W3SVC or WAS, or recycling a pool, hits everyone on the host: flag that and
prefer off-hours.
Verify the failing request returns its real page, the pool stays Started and the right
certificate is served. Note it (PSA Note Discipline base skill): status and sub-status, pool
state, evidence, branch, action or handoff, verification.
Magic Library