Why a version number is not enough
Plants usually record the version number from the application header and stop there. Such a number can be a marketing or bundled label, and one version may include a dozen changes deployed one after another. When a problem appears after an update, the question is not: which version is this, but: after which change did it start?
That is why it is worth agreeing with the vendor on an identifier that points unambiguously to a change. In the system described, the release version in error reports is the commit hash of the deployed image, the same as the image tag in the registry. From one string of characters you can read which change is running on the instance.
Step 1: confirming the version after deployment
The first step after an update is to read the version from the running instance and compare it with the one the vendor announced in the window notice. A match means that exactly this image was deployed, not the previous one. A mismatch is a signal not to proceed to the next steps.
Record the version read in the window protocol. When someone asks in a month which version was in use during an incident, the answer will be in a document, not in someone's memory.
Size of the browser test package
Step 2: a short test of critical scenarios
The full set of browser tests takes about 2 to 2.5 hours, so after a release a short list is used. We base it on the scope that the E2E tests reproduce: logging in, the admin panel, work planning, work permits, reports, documentation, maps and analytics. From this list choose the scenarios on which the safety of the change depends.
The package covers 242 tests in 31 files and runs a real browser against a real backend. A test fails automatically when the browser shows an uncaught JavaScript exception or the server responds with a 5xx status, so a silent error after an update does not go unnoticed.
Pitfall: an error that unit tests do not see
After a fix of a reported bug it is worth running the browser tests, because they catch a user-visible error that unit tests do not detect, since those run against a mocked API. A fix that repairs one function and breaks the flow only shows up in a test on a real backend.
Step 3: error monitoring after the release
For the first hours after a release, watch the error reports split by release version. If a new error type appears only with the hash of the new change, the cause is known: it is that change. Thanks to linking an error to the version, the team decides on a fix or a rollback without a long investigation.
Remember the limits: the limit of events is 21 per hour per error type, so a new, high-frequency error type will be visible as a limited number of events, but with its full description. Work safety alarms are not muted by a cache failure: if the cache fails, the alarm still rings and the log records the problem.
Evidence and release of the operator
The post-release protocol contains: the release version read, the time, the result of the short test of critical scenarios, a reference to the result of the last full run, and the name of the person who released the change to work. It is one page that an internal auditor can compare with previous windows.
The operator returns to work only after the release. This is a simple rule, but its absence is a typical source of problems: the change goes back into a system that has not passed the check, and the first error is reported by a user on the plant floor. The post-release test exists so that the first error is reported by the plant team or the vendor.
- reading the version from the instance and comparing it with the window notice
- a short test of critical scenarios on a real browser
- a hard failure on a JavaScript exception and a 5xx response
- monitoring of error reports split by release version
- a protocol with the version, time, result, and the person releasing the change
Video
Frequently asked questions
Why is the commit hash better than a version number?
It points to exactly one change and is the same as the image tag in the registry. Every error in the reports can be linked to a specific change.
Do all tests have to be run after a release?
No. A full set takes about 2 to 2.5 hours. After a release a short list of critical scenarios is enough, and the result of the last full run serves as a supplement.
What if the version after deployment differs from the announced one?
Do not proceed to the next steps. A difference means that a different image was deployed than agreed, and it has to be clarified with the vendor before the operator is released.
Let's agree on the post-release protocol
We will prepare with you a one-page protocol: version, short test and the person releasing the change. Book a talk with the InnerWeb team.
Let's talk