Twelve Plants, One Data Team: Operating a Fleet of Data Pipelines Without a Night Shift
An illustrative engagement, not a customer story: a manufacturer with twelve sites, a legacy MES at each, and a central team of three. How fleet control, phone approvals, a custom connector and managed operations turn twelve installs into one platform.
- Published
- Reading time
- 8 min
This is an illustrative engagement built from the problems manufacturers bring to us, not a customer story. Revoluzen publishes no customer names until they agree to be named.
The situation
A manufacturer runs twelve plants in three countries. Each plant has a manufacturing execution system from the same vendor, at four different versions, and a local database that must stay on site for latency and for the plant's own operating rules. Head office wants a single view of output, scrap and downtime by shift, and finance wants it reconciled to the ERP.
The central data team is three people. Their current approach is a Python script per plant, run by a scheduler on a server at each site, maintained over remote desktop. The problems are the predictable ones:
- Version drift. Nobody knows which plant runs which version of the script. A fix applied at plant four never reached plant nine.
- Failures nobody sees. When the plant three job fails at two in the morning, the shift report is wrong at seven, and the first person to notice is the plant manager.
- Changes nobody approves. A schedule change at one plant was made directly on the production server by someone who has since left.
- A connector nobody wants to write. The MES has a proprietary export format. Every script re-implements the parsing.
What was delivered
Assess, one week. The plan: one Revoluzen Flow instance per plant, because the data must stay on site, with every instance operated from Revoluzen Console at head office; one custom connector for the MES format, built once and installed everywhere; dev-to-prod promotion so that a change is built centrally, reviewed, and rolled out to all twelve; and managed operations, because three people cannot be on call for twelve sites.
The connector. The MES export becomes a custom connector on the extension layer: parsing, schema mapping, error handling, test coverage, packaged as a versioned extension. It handles all four MES versions. Built once, at the custom connector price, and maintained on the upgrade path: when Flow upgrades, the connector is compatibility-checked against the release.
The fleet. Twelve single-node installs, one per plant, each with the connector and the same four pipelines: output, scrap, downtime, and a reconciliation to the ERP. Console shows all twelve in one view: version, health, last run, failures. Version drift is now visible before it is a problem. A change is promoted from the development environment to all twelve production instances through a review gate; the change shows as a diff, and someone approves it.
The phone. The three engineers get Revoluzen Mobile. A failed run at plant three at two in the morning becomes a push notification, with the failed step attached; the on-call engineer sees the run and the DAG from the phone. A promotion to twelve plants is approved from the phone, with the diff in front of the approver, and the decision is recorded with a name and a time.
Operations. Managed operations, priced per production environment: scheduled health checks, upgrades, backup verification, disaster-recovery tests, incident assistance in business hours with defined escalation. The twelve servers are maintained by someone whose job it is.
Evidence. Each reconciliation run to the ERP emits a signed receipt, so when finance asks why a plant's scrap figure changed, the inputs and the code version are on record.
What it cost
| Item | Price |
|---|---|
| Assessment | fixed, quoted |
| Custom MES connector, four versions | from $1,499 (multi-version, quoted) |
| Verified builds, four pipelines | $149 each |
| Flow Enterprise, Business, twelve production sites | $499 per month per site |
| Managed operations, twelve environments | $399 per month per environment |
That is a predictable monthly number that does not change with production volume, and a central team of three that is no longer a night shift.
The limits, stated
Revoluzen Console and Mobile are in early access; a fleet of this size is exactly the kind of deployment we do with design partners and say so up front. Each plant's instance is self-hosted software the manufacturer owns; managed operations maintains it inside their network. The MES connector is a custom extension and is supported as one.
See the manufacturing page and the services price book.
Frequently asked questions
Why one instance per plant instead of one central instance?
Because the plant data must stay on site, and because a plant should keep running when the link to head office is down. Console gives the central view without centralizing the data.
How are changes rolled out to twelve plants?
Built once in the development environment, promoted through a review gate with a diff and an approval, and applied to the production instances from Console. Rollback goes the same way.
Who handles a failure at two in the morning?
With managed operations, Revoluzen handles routine incidents in business hours with defined escalation; outside those hours the push notification reaches the on-call engineer with the failed step attached. 24/7 coverage is an enterprise agreement.
Does a custom connector lock us in?
No. It is a versioned extension built on published interfaces and delivered with its tests and packaging; it is yours, and it is maintained on the normal upgrade path.
Is this a real customer?
No. It is an illustrative engagement built from the problems manufacturers bring to us.
About the author
Revoluzen
The team building Revoluzen Flow, Catalog, Studio, Console, Mobile and Evidence — writing about what they build and what they have learned running it.