Release Notes — 2.74

Covers the 2.74 release series. Most recent patch: 2.74.3.

Unless a subsection notes otherwise, changes apply to both Monad SaaS and self-hosted.

New features & improvements

SaaS & self-hosted

• Monad MCP — edit one pipeline edge without rebuilding the graph — the new update_pipeline_edge tool changes a single edge's name, conditions, or schema-detection settings in place. Previously an assistant had to resubmit the whole pipeline to touch one edge, and any field it forgot to resend was silently dropped — an omitted condition widened a filtered edge to pass-all, and an omitted schema-detection setting discarded the edge's learned baseline. The tool reads the edge back after the change and reports what actually landed. Docs → (2.74.0) (For teams using the Monad MCP integration.)
• Monad MCP — complete pipeline introspection — get_pipeline now renders each edge's full condition tree (every rule and its configuration, not just the top-level operator), the component each node points at, node names alongside their IDs, and whether a node overrides its component's configuration. An assistant can now audit, document, or faithfully duplicate a pipeline's routing from the MCP alone. (2.74.0) (For teams using the Monad MCP integration.)
• Monad MCP — raw JSON on the pipeline read tools — get_pipeline, get_pipeline_status and get_node_status accept the same raw option the type-lookup tools already had, returning the complete API document when the rendered summary omits a field you need. (2.74.0) (For teams using the Monad MCP integration.)
  • Access: opt-in — pass raw: true on the call; the rendered summary remains the default.
• Monad MCP — forgiving type lookups — the get_input_type, get_output_type, get_transform_type, get_enrichment_type and get_alert_rule_type tools now accept type, id or type_id as well as their canonical parameter, and an unknown type ID returns the closest matches from the catalog instead of an unexplained server error. (2.74.0) (For teams using the Monad MCP integration.)
• Deduplicate and Hash transform operations — two new operations for transforms. Hash writes a digest of the whole record or of selected fields to a key you choose (SHA-256 by default; MD5, SHA-1 and xxh128 are also available), giving you a stable fingerprint for correlation. Deduplicate drops repeat records seen within a rolling window of up to five minutes, keyed on the whole record or on selected fields. Deduplicate docs → · Hash docs → (2.74.2)
  • Access: opt-in — add a Hash or Deduplicate operation to a transform. Deduplicate compares across records, so it must be the only operation in the transform or the last one.

Self-hosted only

• Database migrations now run from the API pod — the separate db-migrations Job is gone; on upgrade the api pod applies migrations from an init container on the same image, so the schema and the code that needs it are always the same release. partition-manager is now its own component, and one global.monad.natsUrl value sets the NATS URL for every component. If your values override anything under db-migrations, remove it; if you run an external Postgres, make sure the monad-db-app Secret includes the uri key. (2.74.0)

Fixes

• Monad MCP status tools answer "is data flowing?" — get_node_status now shows each node's name and type, records in and out, when it last processed a record, and any records shed unattempted; get_pipeline_status's metrics filter works (every value previously returned an error); and sample_node_data now says plainly that it returns a retained sample rather than proof of live flow, pointing to the status tools for liveness. (2.74.0)
• Monad MCP alert and secret guidance corrected — alert-rule severity is now offered as the exact set the platform accepts (the documented warning was never valid and always failed), and the tool descriptions and guidelines show the correct way to reference a secret, which previously led to an unexplained server error. (2.74.0)
• API: partial updates no longer require re-sending the component type — a PATCH to an input, output, or enrichment that omitted the unchanged type field failed with a 400; the stored type is now used, so a description-only or config-only update works as expected. (2.74.0)
• Google Cloud Logs, GKE Audit Logs, and Looker audit logs inputs no longer lose late or backdated entries — the inputs now poll in windows keyed on when Google received each entry rather than its event time, so an entry ingested late is still collected; the Looker input, which had matched no log entries at all, returns logs again. Records arrive a couple of minutes after Google ingests them. Docs → (2.74.0)
• Cisco Meraki Organization Configuration Changes input stays within Meraki's lookback window — a checkpoint older than the 365 days Meraki retains is now clamped forward instead of producing an undefined query, and records without a timestamp no longer fail the run. (2.74.0)
• CrowdStrike Event Stream input resumes correctly after a restart — the stream's application ID is now persisted with the input's state, so a pipeline restart continues the existing feed instead of colliding with it. (2.74.1)
• Lansweeper Vulnerabilities input syncs the full inventory on a schedule — the incremental sync keyed on a global publication date could skip records and never delivered updates to an existing vulnerability; the input now re-lists the full inventory on a Cron schedule you set (required) and drops unchanged records. The former Backfill Start Date setting is removed. Docs → (2.74.0)
• Palo Alto Cloud Security Alerts input no longer skips events after a pause — the relative lookback window is now rounded up, so events from the first minute after a pipeline is re-enabled are collected instead of falling into a gap. (2.74.2)
• Finxact Messages and Error Log inputs no longer stall on a missing detail record — when Finxact returns no detail for a message or error entry, the input emits the entry on its own and moves on instead of retrying the same record indefinitely. (2.74.3)

Breaking changes

None.

Need help?

See the Monad docs for setup guides and reference, or reach out to Monad Customer Support at support@monad.com or via your dedicated Slack customer channel.