DevOps & CI/CD Services in Ghaziabad (2026)

The routine has to be followable by a person who also handles procurement and payroll. We write it as numbered steps with the expected result of each one, so a step that behaves differently is obvious immediately.

The routine is stored where the administrator can reach it without a login to a system that may be down, which in practice means a printed copy in the server room as well as a file.

Most failed upgrades are not caused by the new version being wrong in general; they are caused by it being wrong for this data. Without the previous package and a snapshot from just before the change, the only option is to wait for the vendor.

We keep the last known-good package, a database snapshot and the date each was verified, with a retention that survives a month of inactivity rather than a week.

A test copy of the ERP is only worth having if it holds recognisable data. We refresh it from a recent extract before each planned change, note the date, and restrict access to the people who will actually test.

Where personal or commercial data makes a copy sensitive, the fields that are not needed for testing are masked. That keeps the copy useful without creating a second uncontrolled store of customer records.

The consultant who supports the ERP usually holds an administrator password because the alternative is a support call every month. A named account with a scoped role and an expiry date gives the same convenience with a record attached.

When the engagement ends, access ends on the date written down rather than when somebody remembers. That single practice closes the most common gap found in a small unit's setup.

An upgrade that changes the database is the one that stops a plant. The migration should be rehearsed on the test copy with a realistic volume of data, because a script that runs in four minutes on a sample can take forty on a year of transactions.

We record the rehearsal time and plan the window from it, with a stated point after which the change is abandoned and the previous version restored.

Dispatch hours and shift changes are the two windows to avoid. The calendar is agreed with the plant manager rather than the IT contact, because the plant manager knows when a stopped screen costs the most.

A release that overruns its window is reversed rather than extended. Extending is how a two-hour job becomes a lost shift.

An alert that arrives in a mailbox nobody opens at six in the morning is decoration. The check should call or message the administrator's phone, and it should test the business function rather than the server: can an invoice be raised, can a dispatch be saved.

We keep the alert list short, three or four conditions, so that each one is believed when it fires.

Units change consultants more often than they change software. A handover pack with the package location, the snapshot routine, the configuration values and the three recovery steps means the next person starts from knowledge rather than from an interview.

We keep it to a few pages. A forty-page manual is not read at seven in the morning.

A single application with a saved previous version takes about two weeks. An ERP process with a test copy takes two to three, and two systems sharing a runbook take three.

Handover is a walkthrough with the administrator and one rehearsal of the reversal, performed while the system is quiet rather than during an incident.