v3.24.0

Backward incompatibility

• snow app commands for Snowflake App Runtime projects now always operate on APPLICATION SERVICE objects and no longer fall back to plain SPCS SERVICE objects. snow app open --settings always uses the app-service Snowsight URL segment, and snow app teardown always drops and verifies the app via APPLICATION SERVICE. Apps still deployed as legacy SPCS services are no longer supported by these commands.

Deprecations

New additions

• Added optional --validation-profile to snow spcs service remote-build (e.g. ML_JOB, NOTEBOOK). The value is forwarded as validation_profile on POST /api/v2/remote-build/execute so the image builder can select a pre-baked validation ruleset. When omitted, no image validation is requested. Ignored for --build-type app.
• Added snow spcs service remote-build, remote-build-status, and remote-build-history commands that submit and track SPCS image/app builds via the remote build REST API. These commands are hidden by default and gated by the enable_spcs_remote_build CLI feature flag.
• snow dcm plan now supports the --delta flag for incremental deployments. This option enables processing only statements that have changed since the last deploy, plus statements potentially impacted by those changes.
• snow dcm commands are now generally available (GA). DCM provides infrastructure-as-code capabilities for managing Snowflake objects through declarative SQL files. All DCM commands are now available without needing to enable the enable_snowflake_projects feature flag.
• DCM projects: snow dcm deploy, snow dcm plan, snow dcm preview, and snow dcm raw-analyze now support environment variables and secrets declared in a project manifest's templating.env_vars/templating.env_secrets sections. Declared values are collected from the shell environment, or from a .env file passed via --env-file/-e (the shell always wins for a name declared in both), and forwarded to the server; raw-analyze previously had no support for this at all.
• Added snow helpers check-version, which reports the installed Snowflake CLI version alongside the latest published version and whether an upgrade is available. Use --refresh to bypass the local cache and query PyPI/Homebrew directly. This is the on-demand equivalent of the automatic upgrade banner and always reports its result, even when ignore_new_version_warning is set.
• Added snow dbt copy, which copies files between a local directory and a stage (or between stages) for a dbt project. It is an alias of snow stage copy and supports the same paths and options (--recursive, --overwrite, --parallel, --auto-compress, --refresh).
• snow app events now surfaces app health telemetry for Snowflake App Runtime projects. --type log|metric|lifecycle selects the stream (default log); --since / --until accept relative shorthand (e.g. 30m, 6h, 2d) or absolute UTC timestamps and switch logs to the historical event table, while metric and lifecycle are always historical and default to the last hour. --metric cpu|memory|network narrows metric output and --raw emits unconverted values (bytes, cores). The bare snow app events live log tail is unchanged.
• snow dcm deploy, snow dcm purge, snow dcm plan, snow dcm preview, snow dcm refresh, and snow dcm test now show live, per-step progress as an interactive checklist instead of a single generic spinner.

Fixes and improvements

• snow dcm commands now summarize the files being uploaded as a tree beneath the upload step, instead of listing every file on its own line.
• Recursive stage uploads now upload directories concurrently instead of one at a time. This applies to every command that uploads a directory tree to a stage (e.g. snow stage copy --recursive, snow dcm deploy/plan/analyze, snow dbt deploy, snow spcs service build-image). Upload wall-clock is dominated by per-PUT round-trip latency, so trees with many nested folders (e.g. one file per directory) see a large speedup. Total upload concurrency is bounded by the SNOWFLAKE_CLI_STAGE_UPLOAD_WORKERS setting (config key cli.stage_upload_workers, default 16; set to 1 to restore the previous serial behavior). For snow stage copy this budget is shared with --parallel rather than multiplied by it, so existing invocations do not spawn more threads than before.
• snow dcm plan, deploy, and purge now render each altered object's changes as an indented tree, showing added, modified, and removed columns, constraints, grants, and other properties — with previous → new values — instead of only the object name. Long or multi-line values are collapsed to a single line.
• A bare USER$ database resolved from the active connection is now expanded to the caller's personal database (USER$<username>) when the connection specifies a username, so commands validate and report against the fully-qualified name.
• snow app commands (validate, open, events, deploy, teardown) now resolve their database and schema from the active connection like other CLI commands, so a bare USER$ database configured in snowflake.yml is expanded to the caller's personal database.
• snow app setup now defaults to a workspace for code storage on all databases (previously regular databases used a stage). When the role cannot create a workspace it uses a stage instead, and snow app deploy also falls back to a stage if a workspace cannot be used.
• Upgraded GitPython from 3.1.50 to 3.1.57.
• snow dcm commands run with --save-output now write artifacts directly into the out/ directory, with each command's response in out/<command>_result.json, instead of nesting them under a per-command subdirectory. The out/ directory is recreated empty at the start of every such run.