Snowpipe BCDR on Azure

Overview

This skill helps developers and data engineers design, implement, and operate Snowpipe business continuity / disaster recovery on Azure ADLS Gen2. Six patterns are supported — pick one based on your RPO, RTO, edition, and cost tolerance.

Options 2–6 require Business Critical Edition for Failover Groups. On lower editions, only Option 1 applies.

Decision

Business Critical?
├── No → Option 1
└── Yes
    ├── Zero data loss?           → Option 2
    ├── Azure is the failure?     → Option 4 or 6
    ├── Snowflake is the failure? → Option 5
    ├── Have RA-GRS?              → Option 3
    └── Budget constrained?       → Option 1 or 5

This skill is Azure ADLS Gen2 only. AWS S3 / GCS variants are out of scope.

Core Rules

1. Single writer per file set. Only one pipe active per file set, except Option 2 (which dedups).
2. Storage integration must allow both URLs:

``sql STORAGEALLOWEDLOCATIONS = ( 'azure://<primary>.blob.core.windows.net/<container>/', 'azure://<secondary>.blob.core.windows.net/<container>/' ); ``

1. COPYHISTORY is replicated in Failover Groups but only retains 14 days. For longer history, maintain a FILELOAD_HISTORY backup table.
2. ALTER PIPE ... REFRESH is hard-capped at 7 days. For older files, use DIRECTORY(@stage) + COPY INTO.
3. Inbound notification integrations (DIRECTION = INBOUND) do NOT replicate. After failover-group promotion you must recreate them with the same name in the DR account, re-establish Service Principal trust, and create the DR-region Event Grid subscription.

Workflow

Step 1 — Confirm prerequisites

• Edition (Business Critical for Options 2–6)
• Azure storage type (LRS / GRS / RA-GRS / GZRS / RA-GZRS)
• Event Grid + Storage Queue exists in target regions

Step 2 — Implement chosen option

Each option follows the same shape: storage integration → stage → pipe (with AUTO_INGEST=TRUE and notification integration) → monitoring views → catchup procedure.

Option 2 dedup hash columns:

MD5(METADATA$FILENAME)                          AS FILE_HASH,
MD5(CONCAT_WS('|', $1, $2, $3, $4, $5))         AS RECORD_HASH,
'PRIMARY'                                       AS SOURCE_REGION

Dedup with a Dynamic Table: QUALIFY ROWNUMBER() OVER (PARTITION BY FILEHASH, RECORDHASH ORDER BY LOADTIMESTAMP) = 1.

Option 6 dual pipe pattern:

CREATE PIPE PIPE_PRIMARY   AUTO_INGEST=TRUE INTEGRATION='NOTIF_INT_A' AS
  COPY INTO TARGET_TABLE (..., 'PRIMARY' AS _source_region)   FROM @STAGE_PRIMARY;
CREATE PIPE PIPE_SECONDARY AUTO_INGEST=TRUE INTEGRATION='NOTIF_INT_B' AS
  COPY INTO TARGET_TABLE (..., 'SECONDARY' AS _source_region) FROM @STAGE_SECONDARY;
ALTER PIPE PIPE_SECONDARY SET PIPE_EXECUTION_PAUSED = TRUE;

⚠️ STOPPING POINT: Show planned CREATE STORAGE INTEGRATION, CREATE PIPE, and ALTER FAILOVER GROUP statements to the user and wait for approval before executing.

Step 3 — Validate

SELECT PARSE_JSON(SYSTEM$PIPE_STATUS('pipe_name')):executionState::STRING  AS state,
       PARSE_JSON(SYSTEM$PIPE_STATUS('pipe_name')):pendingFileCount::NUMBER AS pending;

Monitor pending files (>1000 for >15 min), error rate from SNOWFLAKE.ACCOUNTUSAGE.COPYHISTORY (>5%/hr), and load latency (>30 min).

Step 4 — Failover / failback / catchup

ALTER PIPE pipe_primary   SET PIPE_EXECUTION_PAUSED = TRUE;
-- record checkpoint from COPY_HISTORY
ALTER PIPE pipe_secondary SET PIPE_EXECUTION_PAUSED = FALSE;
ALTER PIPE pipe_secondary REFRESH MODIFIED_AFTER = '<checkpoint-15min>';

⚠️ STOPPING POINT: Confirm the failover decision and the checkpoint timestamp with the user before pausing the primary pipe.

Catchup choices: ALTER PIPE … REFRESH (≤7 days), COPY INTO … FROM @stage (any age), or a cursor proc that diffs DIRECTORY(@stage) against COPY_HISTORY for an audit trail.

Common Mistakes

Stopping Points

• Step 2 — wait for approval before running CREATE/ALTER for storage integrations, pipes, and failover groups.
• Step 4 — confirm the failover decision and checkpoint timestamp before pausing the active pipe.

References

• Stage, Pipe, and Load History Replication
• Automating Snowpipe for Azure Blob Storage