Estuary Catalog History

flowctl catalog history - Publication History for Catalog Resources

Concepts: Every change to a catalog resource (capture, materialization, collection) creates a publication. catalog history shows the full timeline of publications including who made the change, what changed, and the full spec at each point in time.

Basic Usage

View publication history for any catalog resource:

flowctl catalog history --name 

Examples:

# Capture history
flowctl catalog history --name acme/source-postgres

# Materialization history
flowctl catalog history --name acme/materialize-snowflake

# Collection history
flowctl catalog history --name acme/orders

With Full Specs (--models)

Include the complete spec at each publication point. Requires json or yaml output:

flowctl catalog history --name acme/source-postgres --models -o json | jq

Output Format

Output is newline-delimited JSON (NDJSON — one JSON object per line). Use jq -s to slurp into an array for filtering/sorting.

Fields per entry:

• catalog_name — resource name
• catalog_type — capture, materialization, collection, etc.
• publication.detail — human-readable change description (e.g. "Published via flowctl", "auto-discover", "periodic publication")
• publication.publicationId — unique publication ID for this entry
• publication.publishedAt — ISO 8601 timestamp
• publication.userEmail — who published
• publication.userFullName — display name of publisher
• publication.userId — UUID of publisher
• publication.model — full spec at that point (only with --models, otherwise null)

Example JSON output:

{"catalog_name":"acme/materialize-snowflake","catalog_type":"materialization","publication":{"detail":"Published via flowctl","model":null,"publicationId":"119fa3bab9000078","publishedAt":"2025-07-01T15:58:30.115433Z","userEmail":"alice@acme.com","userFullName":"Alice Smith","userId":"3be2166e-28a2-440f-876c-a9c2259e413c"}}

Common Detail Values

The publication.detail field can be null, empty string "", or a string. Details are often multi-line (separated by \n) combining a primary reason with specific actions taken.

User-initiated: | Detail | Meaning | |--------|---------| | null or "" | Manual publication via UI (no description) | | Published via flowctl | User published via CLI |

System-initiated: | Detail pattern | Meaning | |----------------|---------| | auto-discover changes (N added, N modified, N removed) | Auto-discover updated bindings | | auto-discover changes (...), and re-creating N collections | Auto-discover with collection re-versioning | | updating inferred schema | Inferred schema updated on a collection | | periodic publication | Scheduled periodic publication | | in response to publication of one or more depencencies | Cascade from upstream change (note: "depencencies" is a typo in the system) | | in response to change in dependencies, prev hash: ..., new hash: ... | Dependency hash changed, materialization rebuilt | | adding binding(s) to match the sourceCapture: [...] | sourceCapture auto-added bindings to materialization | | system created publication in response to discover: | System discover response | | Data_Flow_Reset : disable capture : | Dataflow reset initiated |

Action suffixes (appended after \n to the primary reason): | Suffix | Meaning | |--------|---------| | - updated resource /_meta of N bindings | Resource metadata update | | - updated inferred schema | Inferred schema was actually changed | | - backfilling binding ["schema", "table"] due to incompatible field selection (onIncompatibleSchemaChange: backfill) | Binding backfilled due to schema incompatibility | | - backfilled binding of reset collection | Binding backfilled because its source collection was reset | | and applying onIncompatibleSchemaChange actions | Schema change actions were applied |

Prefix Queries

Use a catalog name prefix to get history across multiple related tasks at once:

# All resources sharing a prefix (capture + its collections)
flowctl catalog history --name acme/source-postgres
# Returns history for acme/source-postgres, acme/source-postgres/table1, etc.

Note: The prefix must be a valid catalog name prefix (no trailing slash). It matches any resource whose name starts with the given string.

jq Recipes

> Important: Output is NDJSON (one object per line). For per-line transforms, jq works directly. For array operations (filtering, sorting), use jq -s to slurp into an array first.

Filter by date range

flowctl catalog history --name acme/source-postgres -o json | \
  jq -s '[.[] | select(.publication.publishedAt >= "2025-01-01" and .publication.publishedAt  /tmp/spec-new.json
flowctl catalog history --name acme/source-postgres --models -o json | \
  jq -s '.[1].publication.model' > /tmp/spec-old.json
diff /tmp/spec-old.json /tmp/spec-new.json

Find auto-discover vs manual publications

# Auto-discover publications (system-initiated)
# Use length > 0 guard before test() to handle empty detail strings safely
flowctl catalog history --name acme/source-postgres -o json | \
  jq -s '[.[] | select(.publication.detail | length > 0 and test("auto-discover|evolved|periodic"))]'

# Manual publications (user-initiated)
flowctl catalog history --name acme/source-postgres -o json | \
  jq -s '[.[] | select(.publication.detail | length == 0 or test("auto-discover|evolved|periodic") | not)]'

Real-World Investigation Patterns

1. Unexpected Backfill Investigation

A materialization backfilled unexpectedly — find the triggering change:

# Find backfill-related publications on the materialization
flowctl catalog history --name acme/materialize-snowflake -o json | \
  jq 'select(.publication.detail | length > 0 and test("backfill|incompatible")) | {at: .publication.publishedAt, detail: .publication.detail}'

# Check the source collection for schema changes around that timestamp
flowctl catalog history --name acme/orders -o json | \
  jq -s '[.[] | select(.publication.publishedAt >= "2025-01-15" and .publication.publishedAt  0 and test("auto-discover")) | {at: .publication.publishedAt, detail: .publication.detail}'

Common root causes: auto-discover detected type change (VARIANT → TEXT), field type widened (integer → string), manual destination table changes.

2. Auto-Discover Changed Something Unexpectedly

Auto-discover added/removed bindings or changed config without user intent:

# Find all auto-discover publications
flowctl catalog history --name acme/source-postgres -o json | \
  jq 'select(.publication.detail | length > 0 and test("auto-discover"))'

# Use --models to diff what changed between the auto-discover and previous publication
flowctl catalog history --name acme/source-postgres --models -o json | \
  jq -s '.[0].publication.model' > /tmp/after.json && \
flowctl catalog history --name acme/source-postgres --models -o json | \
  jq -s '.[1].publication.model' > /tmp/before.json && \
diff /tmp/before.json /tmp/after.json

Common root causes: connector returned incomplete discover response, resource paths changed, access revoked causing bindings to be removed.

3. Who Changed What and When (Audit Trail)

Need to understand who modified a connector and when:

# Full audit trail with user attribution
flowctl catalog history --name acme/source-postgres -o json | \
  jq '{at: .publication.publishedAt, who: .publication.userFullName, email: .publication.userEmail, detail: .publication.detail}'

# Find all changes by a specific user
flowctl catalog history --name acme/source-postgres -o json | \
  jq -s '[.[] | select(.publication.userEmail == "alice@acme.com")]'

# Filter OUT system changes to see only human-initiated publications
# Note: use "== ... | not" instead of "!=" — zsh escapes ! in double quotes
flowctl catalog history --name acme/source-postgres -o json | \
  jq -s '[.[] | select(.publication.userEmail == "support@estuary.dev" | not)]'

Note: publication.userFullName can be null for some users — always show userEmail as a fallback.

4. Endpoint Config / Credential Changes

Server address, credentials, or database settings changed:

# Use --models to compare config across publications
flowctl catalog history --name acme/source-postgres --models -o json | \
  jq -s '.[] | {at: .publication.publishedAt, who: .publication.userEmail, detail: .publication.detail}'
# Then diff individual models to find the specific config change

Common root causes: server address changed (prod → dev), credentials rotated, database name changed.

5. Binding State Changes (Disable/Enable/Remove)

Bindings were disabled, removed, or re-enabled unexpectedly:

# Find publications mentioning binding changes
flowctl catalog history --name acme/materialize-snowflake -o json | \
  jq 'select(.publication.detail | length > 0 and test("disable|enable|remove|binding"))'

# Compare binding counts across publications with --models
flowctl catalog history --name acme/source-postgres --models -o json | \
  jq '{at: .publication.publishedAt, detail: .publication.detail, bindings: (.publication.model.bindings | length)}'

Common root causes: onIncompatibleSchemaChange: disableTask auto-disabling, auto-discover removing bindings when connector returned incomplete response, user lost database access.

Cross-Task Timeline Reconstruction

Example workflow: "Why did my materialization backfill?"

1. catalog history on materialization → find backfill publication timestamp and publicationId
2. catalog history on source collection → find schema change around same timestamp
3. catalog history on capture → find auto-discover that triggered the schema change
4. Correlate timestamps: auto-discover → collection schema update → materialization backfill

When to Use

• Investigating what changed and when for a catalog resource
• Finding who made a specific change (audit trail)
• Correlating changes across capture → collection → materialization chains
• Debugging unexpected backfills by tracing the publication timeline
• Comparing specs between publications to see exactly what changed
• Auditing auto-discover vs manual changes
• Investigating binding state changes (disable/enable/remove)
• Understanding endpoint config or credential changes

Related Skills

• estuary-catalog-status — Check current status and runtime state (also has controller_status.publications.history in its JSON output)

Tips

• Output is NDJSON (one JSON object per line) — use jq -s to slurp into an array for filtering/sorting
• Without -s, jq processes each line individually (useful for simple transforms like jq '{at: .publication.publishedAt, detail: .publication.detail}')
• Default output format is YAML; use -o json for scripting
• publication.detail can be null, "", or a string — always guard test() with length > 0 to handle both safely
• zsh gotcha: != in jq gets escaped by zsh (\!=). Use == "value" | not instead
• Prefix queries are powerful for seeing the timeline across related resources
• Combine with flowctl catalog status to see both current state and change history

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: estuary
• Source: estuary/agent-skills
• License: Apache-2.0
• Homepage: https://estuary.dev

Install and usage instructions live in the source repository linked above.