Nothing was down. The invoice was the fire: close to $700k a year in Alation licensing for roughly 1,300
users, and an enterprise vendor search already running to replace it — with another vendor. It just
doesn’t make a whole lot of sense to keep spending seven hundred thousand dollars a year for the users we
have. The features that justified the price were already owned: Collibra held the business metadata, other
in-house systems covered the rest. The enterprise was paying twice for capability and once for a query
editor.
Fig. 01 — Arithmetic, not an invoice. Close to $700,000 a year
across roughly 1,300 users is about $540 a seat, and what the seat bought was a query editor.
No feature-parity option existed. DBeaver doesn’t do most of what Alation does, and rebuilding all of it
was never on the table — the swap only worked because those features were already duplicated in tools the
enterprise owned. The business didn’t really care. The databases declare almost no foreign keys, so any
join help had to be mined, not read. And roughly 1,300 users had to move without a stop-the-world cutover.
Fork open-source DBeaver instead of buying the next vendor. The search was already
headed toward a like-for-like replacement. Rejected it, because the differentiating features were
already owned elsewhere — the seat being paid for was, in practice, a query editor. DBeaver’s Apache 2
license made the fork legal and clean.
A hard fork, not plugins. The changes went too deep for the extension surface. The cost
is accepted with eyes open: nothing goes back upstream, and the fork is maintained in-house from now on.
Wire in the metadata the enterprise already owned rather than rebuild a catalog.
Collibra’s business metadata streams into IntelliSense — as the completion list appears, the developer
sees in real time what each column means. Rejected: replicating Alation’s catalog inside the new tool.
Mine joins instead of trusting the schema. With no declared foreign keys, scanned every
stored procedure and view, extracted the join predicates into their own table, and surfaced them
whenever a user writes a join.
Not again: don’t keep renewing a vendor whose differentiating features the enterprise
already owns.
Two to three months from vendor comparison to production. Scheduled queries run on Temporal — write a
query in the tool, schedule it for seven in the morning, results go out by email or land in S3. Everyone
gets schema-level read access by default — every database, every table, every column visible; touching
data requires your own credentials. Alation is sunsetting over the coming months, and the licensing spend
goes with it when the contract ends. No adoption metric published.
The query tool is owned in-house now — a fork with no route back to upstream, maintained by the enterprise
that runs it. That trade was made knowingly. The renewal conversation is over.
The client ships by name; the people do not. No internal org names, no team names. The vendors are the
story and stay named: Alation out, DBeaver forked. User count rounded; the licensing figure is as it was
reported internally. Stack footnote: DBeaver (Apache 2), Collibra, Temporal, ECS Fargate, Aurora
Serverless v2, ALB with ACM, Amazon S3.
