skip to content

Blocking a Bad Change

A gate is a control with an owner, an authority and a bypass, deciding on evidence some other tool produced. Interviewers probe it because a gate everyone routes around enforces nothing at all.

on this pageshow

explore

questions

page 1 of 2

What is a break-glass override of a blocking policy gate, and what must it leave behind?

level: juniorimportance: must knowfreq 66%

answer

  1. a planned exit, not a hole
  2. the bypass happens either way
  3. one change, not the whole rule
  4. loudness is the safeguard
  5. who, what, when, which rule

basics

~20 s

A break-glass override is a pre-authorised, deliberately loud way to ship one change that a policy gate denied, used when waiting is worse than the risk. It must leave a record: who pulled it, which rule, which change, and why.

solid answer

~40 s

A blocking gate that offers no legitimate way past it still gets passed - someone merges with elevated rights, edits the check, or deploys outside the pipeline. Break-glass makes that path explicit and observable instead of improvised. Three properties define it: it is authorised in advance so nobody invents it under pressure, it is scoped to a single change rather than to the rule, and it is recorded and alerted at the moment it is used. The record is the whole point - an override that leaves no trace is indistinguishable from a hole in the gate, and the deterrent is that someone else sees it, not that it is hard to do. It bypasses the gate, not the scrutiny: the change still gets reviewed, just afterwards.

go deeper

for a junior

Be ready to define it in one sentence and to say that a bypass must be recorded and attributable to a person. Knowing that it applies to one change rather than switching the rule off is enough at this level.

for a middle

An interviewer expects you to explain why a blocking gate needs a sanctioned bypass at all - that the bypass happens regardless, and the only choice is whether it happens inside the system. Contrast it with downgrading the rule to warn-only.

for a senior

Show that you would wire the alert to a third party and make the effect expire, and that you can tell an emergency from an inconvenience without a rulebook. Name the after-the-fact review as part of the mechanism, not an optional extra.

for a principal

Own the position that break-glass exists at all, and be able to defend it to someone who reads a bypass as a broken control. The argument is about attribution replacing false certainty, and about who is accountable when the pull rate climbs.

## The problem break-glass solves A policy gate that blocks turns a rule into a hard stop: the change does not merge, or the deploy does not run, until the rule is satisfied. That is the point of a blocking gate, and most of the time the answer "fix the change" is correct. But some fraction of denials arrive at a moment when fixing the change properly costs more than the risk the rule was written to prevent - a service is down, a customer-facing failure is spreading, and the correct fix happens to violate a rule. An organisation has exactly two options at that moment. Either there is a sanctioned way through, or people invent one. The invented ones are worse in every respect: an administrative merge, a hand-run deploy from a laptop, a quick edit that makes the check pass without fixing anything. All of them ship the change, and none of them tell you afterwards that a rule was violated. Break-glass exists because the bypass is going to happen; the only question is whether it happens inside the system or outside it. ## What makes it a break-glass override rather than a hole **It is authorised in advance.** Somebody decided, in daylight, that this gate has an emergency path and who is allowed to use it. Nothing is invented at 02:00 except the decision to use it. **It is scoped to one change.** Break-glass lets a specific change past a specific rule. It does not switch the rule off, and it does not apply to whatever else deploys that night. A bypass whose blast radius is "every pipeline until someone remembers" is a rule change wearing an override's clothes. **It is recorded and it is loud.** The pull produces an attributable record at the time it happens - who, which rule, which change, when, and the reason given - and it notifies somebody other than the person who pulled it. The exact fields the record must carry and how long it is kept are a compliance concern; what matters here is that the record exists at the moment of use and names a person, because a reconstruction assembled a week later from chat scrollback is not a record. **It is temporary.** The effect ends with the change. Whatever was loosened either expires, is reverted, or becomes a deliberate rule change through the normal path. ## What it is not It is not the same as switching the rule from blocking to warning. Downgrading a rule changes the default for everyone and keeps changing it until someone flips it back; a break-glass pull spends a single, visible exception and leaves the default deny intact for the next change. It is also not the same as a planned exception granted in advance for a known situation. A planned exception is negotiated with time to think and covers a class of changes. Break-glass is the unplanned case: no time, one change, and the record substitutes for the review that could not happen first. And it is not an admission that the gate is fake. Interviewers probe this, and the answer worth giving is that the deterrent in a bypassable gate is attribution, not difficulty. The rule still states what the organisation wants, the gate still stops the accidental case by default, and the override converts a silent violation into one with a name on it. ## The failure modes worth naming **The silent override.** A skip flag exists, everyone knows it, nothing alerts when it is used. This is the most common failure, and it is worse than having no gate because it produces a green history that nobody can trust. **The override that stops expiring.** Whatever was loosened for one deploy stays loosened, because reverting it was nobody's task once the incident closed. **The override that becomes the workflow.** If a team reaches for break-glass every week, the pull has stopped being an alarm. That is a signal about the rule or the process, and it needs to be read rather than tolerated. **The shared handle.** A bypass performed through a shared account or an unattributed token records that "someone" overrode the rule, which is the one thing the record was supposed to prevent. ## What to say in an interview Define it in one sentence, say why a blocking gate needs one at all, and then name the record and the alert as the parts that make it a control rather than a hole. If you can add that the follow-up review is mandatory and owned by a person, you have answered above the level the question is usually asked at.

  • If anyone can trigger the override, is the gate worth having at all?
    Yes. The gate still states the rule and still stops the accidental case by default - most denials are honest mistakes that get fixed. What the override changes is the small remainder: instead of someone finding an unmonitored way around, the violation ships with a name, a reason and an alert attached. The deterrent is visibility, not difficulty.
  • How does a break-glass override differ from switching the rule to warn-only?
    Switching to warn changes the default for every change and every team, and it stays changed until somebody remembers to flip it back. A break-glass pull leaves the default deny intact and spends one visible exception on one change. If your response to a blocked emergency is to downgrade the rule, you have permanently traded away the control to solve a five-minute problem.
  • Who should be notified when someone uses the break-glass path?
    Someone other than the person who used it, within the same shift - typically the rule's owner and whoever is on call for security. A notification that lands in a monthly report is not an alert; it is filing. The point of notifying a third party is that the pull is observed while the context is still recoverable.

It is the glass on a fire alarm: pulling it is allowed and sometimes right, but it breaks visibly, everyone hears it, and someone comes to ask what happened.

saying these in an interview costs you the question

  • Says break-glass means turning the rule off for the day
  • Thinks urgency removes the need for a record
  • Claims a gate that can be bypassed is not a real control
  • Treats a verbal approval in a chat channel as the record
  • Confuses break-glass with a pre-agreed standing exception

context

open as a page

A CI check rejects plans requesting a disallowed instance type, so why do such instances still appear in the account?

level: juniorimportance: must knowfreq 70%

basics

~20 s

A CI check only sees changes that travel through CI. Anyone holding cloud credentials can create the instance from a console session or a local apply, and that route never reaches the check at all.

open as a page

Why is a pull-request policy check advisory when an in-cluster reconciler is the only thing that applies manifests?

level: juniorimportance: must knowfreq 60%

basics

~20 s

Merging is not deploying. The reconciler reads whatever sits on the tracked branch and applies it, without consulting the check that ran on the pull request. The check informs humans; it has no authority over the applier.

open as a page

In a pull request, what is a required check, and what happens if it never reports a result?

level: juniorimportance: must knowfreq 68%

basics

~20 s

A required check must report a passing result before the pull request may merge. An advisory check only posts a result nobody consults. If a required check never reports, the merge stays blocked rather than allowed.

open as a page

A build gate finds no vulnerability scan report attached to an artifact — is that a pass?

level: juniorimportance: must knowfreq 70%

basics

~20 s

No. A missing report is not a clean result, it is an absent one. A gate needs three outcomes — pass, fail, and no usable evidence — and the third must never be folded silently into the first.

open as a page

What does a rule requiring cost-centre and on-call fields in a service catalog entry actually check?

level: juniorimportance: must knowfreq 70%

basics

~20 s

A required-metadata rule checks only that named fields exist in the catalog entry and hold a value from an allowed set. It proves the entry is well-formed and names an owner, not that the values are true.

open as a page

Why can't a release gate trust a scan report that the build job wrote about itself?

level: juniorimportance: must knowfreq 62%

basics

~20 s

The build job is the thing being gated, so a report it wrote about itself is a claim, not a check. Anything that skips, misconfigures or compromises the build can also produce a passing file.

open as a page

In a CI pipeline, why does a green policy check not prove the gate actually evaluated that commit?

level: juniorimportance: must knowfreq 55%

basics

~20 s

Green means nothing failed, not that something ran. A skipped job, a path filter or a check nobody required all look successful. Without a record asserting an evaluation happened, passed and never invoked are indistinguishable.

open as a page

A CI policy check times out and the pipeline records it as passed. Why is that unsafe?

level: juniorimportance: must knowfreq 62%

basics

~20 s

A timeout means the rule produced no decision at all, not that the change is compliant. Recording it as a pass turns an overload into a silent approval, and that happens most often exactly when the system is busiest.

open as a page

What is a severity threshold in a pipeline security gate, and what does it actually decide?

level: juniorimportance: must knowfreq 72%

basics

~20 s

A severity threshold is the rule that collapses a scan report into pass or fail: block when a finding sits at or above a chosen severity, such as critical. It decides what stops a change, not what is genuinely risky.

open as a page

The same rule runs as a PR check, a deploy job step and an admission controller — which still applies to a manual change?

level: middleimportance: must knowfreq 62%

basics

~20 s

Only the admission controller, and only for objects sent to the cluster API it fronts. A pull request check and a deploy job step both live inside a pipeline, so an engineer applying the change directly skips both.

open as a page

A policy rule that calls a registry and live API discovery fails one run in ten. How do you make it deterministic?

level: middleimportance: must knowfreq 54%

basics

~20 s

Split acquiring the external data from deciding on it. One step fetches and pins the registry and discovery facts, the rule then evaluates that pinned document offline, and a failed fetch is reported as an error, never as a verdict.

open as a page

In a security findings gate, how does blocking on new findings differ from blocking on total findings?

level: middleimportance: must knowfreq 60%

basics

~20 s

Blocking on total findings fails every build until the entire backlog is clean. Blocking on new findings compares each report against a recorded baseline of what already existed and fails only on what the current change added.

open as a page

The only proof your encryption-at-rest gate passed is a file in the gated team's own repo. What does it prove?

level: seniorimportance: must knowfreq 45%

basics

~20 s

That someone committed a file saying pass. A record the gated party can write, edit, regenerate or delete carries almost no weight about whether the gate ran, and a missing file cannot be told apart from a deleted one.

open as a page

When you design a break-glass path around a blocking gate, should the override cost a click or a ticket?

level: middleimportance: should knowfreq 50%

basics

~20 s

Make the override cheap to pull and expensive to hide. A ticket-priced bypass nobody can reach at 02:00 gets routed around; a one-click bypass that names the user, alerts others and covers one change keeps the event visible.

open as a page

Under pull-based delivery, a merged manifest is denied by cluster policy - where does that denial surface, and who sees it?

level: middleimportance: should knowfreq 48%

basics

~20 s

It surfaces on the reconciler: a failed sync condition and a degraded or unhealthy status carrying the rejection message, plus the agent logs. It never reaches the pull request, which closed green before the apply was attempted.

open as a page

Why can a pull request that edits the pipeline file stop its own security gate from running?

level: middleimportance: should knowfreq 55%

basics

~20 s

Most CI systems read the pipeline definition from the commit under review, so one change can disable, rename or narrow the very job that would have judged it — and can equally edit a policy file vendored beside the code the rule inspects.

open as a page

Why should a policy rule read normalised finding fields rather than each scanner's native output?

level: middleimportance: should knowfreq 55%

basics

~20 s

Because a rule bound to one tool's schema must be rewritten for every other tool and breaks when that schema changes. A normaliser maps each report into one record shape, so policy is written once against stable fields.

open as a page

A licence-class rule must block copyleft in a shipped binary yet allow it internally - how does the rule tell the two apart?

level: middleimportance: should knowfreq 45%

basics

~20 s

The rule reads two inputs: licence class per component from the build's inventory, and the service's distribution context from its catalog entry. Applicability comes from the catalog, so one component passes internally and fails in a shipped binary.

open as a page

Why must the expected builder identity live in the gate's rule rather than be read from the evidence?

level: middleimportance: should knowfreq 47%

basics

~20 s

A field read out of the evidence says whatever its producer wanted it to say. Pinning the accepted builder identity in the rule is the step that compares the claim against a value the gated team cannot edit.

open as a page

As the approver holding break-glass authority at 02:00, how do you decide whether to override an egress rule blocking an outage fix?

level: seniorimportance: should knowfreq 60%

basics

~20 s

Decide whether the rule is right about this change, not whether the incident is urgent. Grant the narrowest override that restores service, record it as you grant it, and book the review that either reverts the loosening or changes the rule.

open as a page

Resources keep appearing that your pipeline policy would have rejected — how do you make the pipeline the only path?

level: seniorimportance: should knowfreq 48%

basics

~20 s

Stop adding checks and start closing routes. Inventory every identity that can write to the surface, reduce standing write access so the deploy identity is effectively the only one, and put detection on whatever routes you choose to leave open.

open as a page

In a pull-based deploy every apply arrives as the reconciler - how do you enforce a rule that depends on who authored the change?

level: seniorimportance: should knowfreq 42%

basics

~20 s

Not at the cluster gate. Every change from every team arrives under the same reconciler identity, so a rule keyed on the requester either allows everyone or denies everyone. Enforce who-rules in the repository, where a human identity exists.

open as a page

How do you design a policy gate that the team it blocks cannot skip or edit away?

level: seniorimportance: should knowfreq 44%

basics

~10 s

Resolve the rule and the decision from somewhere the gated change cannot rewrite, record the requirement outside the repository content, and make a missing or errored decision block the merge instead of passing silently.

open as a page

How do you make an approved-base-image rule match when each tool names the image differently?

level: seniorimportance: should knowfreq 44%

basics

~20 s

Join on the image digest. A tag is a mutable pointer and a package URL is a naming convention, so resolve every report's reference to a digest when the evidence is produced, and treat an unresolvable reference as unknown rather than approved.

open as a page

A repo's scaffold stamp says paved-road template 3.2.0 but the files were later rewritten - what does gating on that stamp prove?

level: seniorimportance: should knowfreq 50%

basics

~20 s

A scaffold stamp records origin, not current state: this repo was created from that template version. Later edits never change it, so gating on the stamp certifies history and says nothing about whether the files still conform.

open as a page

Your gate reads both build-written reports and verified signed statements — how do you set outcomes per class?

level: seniorimportance: should knowfreq 40%

basics

~20 s

Tier the inputs by who could have produced them. Evidence the gated job controls may only warn; evidence from a producer the gated team cannot write to, attributed to an identity policy pinned, may block promotion.

open as a page

The policy gate is your incident: jobs are queuing and every retry goes green on the third attempt. What do you do in the first hour?

level: seniorimportance: should knowfreq 46%

basics

~20 s

Make an explicit, time-boxed decision instead of letting timeouts make it. Use the gate's telemetry to find the slow rule and its dependency, narrow that one rule under a named owner and expiry, and record what shipped un-evaluated.

open as a page

Your zero-criticals gate blocked a one-line change over a pre-existing finding. What do you change?

level: seniorimportance: should knowfreq 54%

basics

~20 s

Rescope the rule from the absolute state to what the change introduced: fail on findings absent from a recorded baseline, and move the pre-existing critical into owned, dated backlog work. Do not simply raise the threshold until the pain stops.

open as a page

A record says your audit-log retention rule passed six months ago. What must it capture to still mean anything?

level: middleimportance: nice to knowfreq 32%

basics

~20 s

The identity of what was judged and what judged it: the input the engine actually saw, the rule set version in force, and an explicit decision. Passed means little unless you know which rule passed it.

open as a page

showing 1–30 of 36