Collibra Instance Health Checks

Prev Next

What this pack contains

Fifteen tests that read a Collibra instance and tell you whether it is working and well governed. They run on a data source built from the Collibra Data Source Template. Twelve are standard tests. Three are template tests that create one child test for every database registered in Collibra, taken from the data source's catalog, so a newly registered database gets its own synchronization, ownership and description checks after the next metadata ingestion.

Every test works the same way. It counts the issues it finds, returns that count, and passes at zero. A failing test prints the list of issues in its result and, where the runtime supports it, attaches the list as a CSV file, so you see the unhealthy Edge site, the table with no steward or the obsolete table still feeding another rather than only a number. Thresholds sit in a PARAMETERS block at the top of each script.

Test Type Dimension Severity Fails when
API reachable and credentials valid Standard Performance Critical Collibra does not answer within the time limit. Refused credentials make the test error. Every other check depends on this one
Edge sites healthy Standard Processing High An Edge site's status is not HEALTHY. Edge runs schema synchronization, profiling and lineage harvesting, so Collibra stops learning about the databases behind an unhealthy site
No failed jobs Standard Processing High A job (schema synchronization, import or classification) failed or completed with errors in the last 7 days
Metadata synchronization current: database Template, per database Timeliness High The database's schemas were last synchronized more than 7 days ago, or never, so Collibra may describe tables that no longer look that way
Table owner and steward coverage: database Template, per database Completeness Medium Fewer than 50% of the database's tables and views have an Owner or a Steward; lists those without one
Responsible users are active Standard Relationships High A user who holds a responsibility (Owner, Steward or Custodian) is disabled, so workflows and questions go nowhere
Description coverage: database Template, per database Completeness Low Fewer than 50% of the database's tables and views have a description; lists those without one
Business terms have definitions Standard Completeness Medium A business term has no Definition, so it cannot settle what a report's numbers mean
No duplicate business terms Standard Uniqueness Medium Two terms in a domain share a name once case, spacing and punctuation are ignored, such as "Order" and "order"
Obsolete tables have no downstream consumers Standard Relationships High A table or view whose status is Obsolete or Missing from source still feeds other columns in Collibra's lineage
No workflows in error Standard Processing Medium A workflow instance is stuck in error, which leaves its asset in limbo
No overdue workflow tasks Standard Timeliness Low An open workflow task is past its due date
No near-duplicate tags Standard Consistency Low Tags differ only in case, spacing or punctuation, such as PII and pii, so searches and filters on one miss the other
Data class suggestions reviewed Standard Security Medium A data classification suggestion has waited more than 14 days for a steward to accept or reject it. An unreviewed suggestion on a column of personal data means the column is not yet treated as sensitive
Validatar feed into Collibra is current Standard Timeliness High Validatar's Data Quality Rules in the Validatar domain have not changed in 2 days

Every test name in Validatar starts with Collibra -, for example Collibra - No failed jobs. Child tests add the database's table name, such as Collibra - Metadata synchronization current: SAMPLE_DW [98512984], and are placed in a folder named after their template test.

The last test watches the integration itself. If the publish step stops, Collibra keeps showing old scores, and this test is how you find out.

A check the integration's Collibra credentials may not run shows as Error, with a message naming what could not be read, rather than passing. Three checks see more with broader roles: Responsible users are active needs a role that can list users, and No failed jobs and No overdue workflow tasks see only the integration's own jobs and tasks unless its role can see every user's.

Where to get it

Marketplace: Collibra Instance Health Checks

Prerequisites

  • A Collibra data source created from the Collibra Data Source Template, with Ingest Technical Metadata run at least once. The template tests select their databases from its Database schema.
  • Collibra credentials that can read the physical data dictionary domains of your databases, and the broader roles above for the checks that need them.
  • To try the pack without Collibra, turn on Demo Mode on the data source. The built-in sample instance makes 13 of the 21 results fail on purpose, so you can see what a failing result looks like.

Import

  1. Ingest the Collibra data source's metadata first. The template tests need its catalog.
  2. Create a folder for the pack in the target project, for example Collibra Instance Health Checks. The tests are imported straight into the folder you choose.
  3. Open Marketplace in Validatar, find Collibra Instance Health Checks, choose Get, then Import Directly.
  4. Choose the target project when prompted. The pack is project content, so Validatar asks which project to put it in.
  5. Map the pack's data source, Collibra (Dev), to your Collibra data source, and choose the folder from step 2.
  6. Choose Review Changes. The review lists the 15 tests to add. Choose Import.
  7. Open each of the three template tests and generate their child tests, so they pick up the databases in your catalog. See Materializing Child Tests.
  8. Run the folder, or add it to a job. Daily is a sensible schedule.

To import from a file instead, download the pack from the marketplace listing and import it on the project's Import page.

Step 2 of the pack import: map the Collibra data source and choose the folder

Adjusting a test

Each script starts with its thresholds. For example, the synchronization check:

# Collibra Instance Health Check: Metadata synchronization current
# PARAMETERS --------------------------------------------------------
MAX_AGE_DAYS = 7           # oldest acceptable synchronization
# -------------------------------------------------------------------

Change a value and save. For a template test, change it on the template test and generate the child tests again. The checking logic lives in the Collibra template, so a newer template version improves every test without re-importing the pack.

Test Parameters (default)
API reachable and credentials valid MAX_SECONDS = 10: slowest acceptable answer
Edge sites healthy SITE_NAMES = "": a comma-separated list of the sites to watch, blank for every site
No failed jobs DAYS = 7: how far back to look
Metadata synchronization current MAX_AGE_DAYS = 7
Table owner and steward coverage MIN_COVERED_PCT = 50
Description coverage MIN_DESCRIBED_PCT = 50
Obsolete tables have no downstream consumers STATUSES = "Obsolete,Missing from source"
No overdue workflow tasks GRACE_DAYS = 0: days past due before a task counts
Data class suggestions reviewed MAX_AGE_DAYS = 14: oldest acceptable open suggestion
Validatar feed into Collibra is current MAX_AGE_DAYS = 2; DOMAIN_ID = "", blank for the Validatar Data Quality domain
Responsible users are active, Business terms have definitions, No duplicate business terms, No workflows in error, No near-duplicate tags none

Metadata synchronization current reads the newest successful job named after the database's Edge connection, and otherwise the newest change to its schema assets. How Collibra names schema synchronization jobs has not yet been confirmed on a live instance.

What the demo instance shows

With Demo Mode on, the sample instance is a fictional company with three databases registered through Edge and these deliberate problems:

Test Result Why
Edge sites healthy Fails The site edge-dmz is UNHEALTHY, with an upgrade pending
No failed jobs Fails The CRM schema synchronization failed two days ago: connection refused
Metadata synchronization current Fails for crm The CRM database was last synchronized 19 days ago
Table owner and steward coverage Fails for crm Two of the three CRM tables have no Owner or Steward
Responsible users are active Fails The steward of the CRM opportunities table is a disabled user
Description coverage Fails for crm Two of the three CRM tables have no description
Business terms have definitions Fails Net Revenue and Churn have no definition
No duplicate business terms Fails "Order" and "order"
Obsolete tables have no downstream consumers Fails STAGE.RAW_ORDERS, status Obsolete, still feeds PUBLIC.ORDERS
No workflows in error Fails An approval workflow on Net Revenue failed to notify its steward
No overdue workflow tasks Fails "Review term: Churn" is 20 days past due
No near-duplicate tags Fails "PII" and "pii", "Customer Data" and "customer-data"
Data class suggestions reviewed Fails Two suggestions on CRM columns have waited more than a month

Everything else passes, including Validatar feed into Collibra is current, because the sample's Validatar domain holds two rules published a day earlier. With the template tests' child tests (one per database) a demo run gives 21 results: 8 pass and 13 fail by design. The sample's SAMPLE_DW database passes every per-database check; it is the one the Collibra Metadata and Lineage Scripts use in their own demo.

The Collibra Instance Health Checks import completed

Troubleshooting

Symptom Cause Fix
A template test has no child tests The Collibra data source's metadata has not been ingested, or the child tests were not generated Run Ingest Technical Metadata, then generate the child tests
A test shows Error saying the credentials may not read something The Collibra identity lacks a permission or role Add it, or remove the test
No failed jobs or No overdue workflow tasks passes while you know of failures The integration's role sees only its own jobs and tasks Give it a role that sees every user's
A child test errors because its database is not found The database was removed from Collibra, or the credentials no longer see its domain Ingest the Collibra data source's metadata again and regenerate the child tests
Metadata synchronization current fails for a database you know is current Its synchronization jobs are named differently from the Edge connection Check the job names in the data source's Job schema; the test falls back to the newest change to the database's schemas
Validatar feed test fails Publishing to Collibra is not set up, is in Dry Run, or uses another domain Set up publishing, or set DOMAIN_ID

Related

Note: These tests are built from Collibra's documentation and published OpenAPI specifications, and have been tested offline against the built-in sample Collibra instance. Inside Validatar, the pack imported from the Validatar marketplace and gave its 21 designed results against that instance: 8 pass and 13 fail. They have not yet been run against a live Collibra instance, so thresholds and some checks may change after that validation.