Skip to content

Check the Dify plugin in a native application - #946

Merged
zlareb1 merged 9 commits into
mainfrom
test/dify-native-application
Nov 21, 2026
Merged

zlareb1 merged 9 commits into
mainfrom
test/dify-native-application

Conversation

@zlareb1

@zlareb1 zlareb1 commented Nov 19, 2026 •

Copy link
Copy Markdown
Member

Changelog category (leave one):

  • CI Fix or Improvement

Changelog entry

Dify tools have a hosted Community Edition application check for registration, workflow execution and stored user identity.

Why

Archive and isolated tool tests do not exercise application setup, plugin discovery or workflow parameter binding. A disposable Community Edition application checks those boundaries against the candidate plugin, server and SDK wheel.

Scope and risk

The application uses current Community Edition pinned to its immutable source commit, ordinary setup/login/workspace debugging routes, signature verification and the configured plugin-daemon image user. The tools use Loom memory and document search; the host's optional dataset subsystem stays disabled through its supported empty vector-store selector. The check requires that selection in the resolved API, worker and Beat configuration and the exact 14-service closure.

The platform's official initialization service owns the fresh plugin-storage bind. Both readiness checkpoints require the original directory identity, owner and initialization marker. No custom root helper or replacement backend runs. Workflow start and end nodes use their declared schemas.

Remember, same-user recall, empty recall for a second user, namespace-shared document search and exactly one seven-field stored identity/value row must pass. The stored fact is compared as the exact JSON text written by memory.set. Image IDs, resource caps and owned process/container/volume/network cleanup remain enforced. Incomplete census, monitor failure or unverified cleanup invalidates success; credentials are redacted from retained evidence.

The final package's README describes the exercised scope. Cloud, browser, protected-session families and host dataset retrieval are untested. The scenario uses fake providers and a candidate-built isolated SDK wheel; it does not qualify the published registry wheel or a complete host/version/architecture matrix. Private runtime files are retained, and intentional interruption was not exercised by the successful application run.

Compatibility, rollout and rollback

No product operation, wire field, setting, schema, migration or runtime behavior changes. The reach manifest records fixture calls. No operator steps are needed; the stack exists only on the disposable hosted runner. Revert the check to roll it back.

Tests

  • The final hosted Dify workflow passes both package and native application jobs. All four workflows, the exact stored row, both storage receipts and owned resource cleanup pass. All 19 archive members match the candidate source, including the packaged README.
  • All 34 ownership/failure guards and 53 subtests pass. Causal regressions cover the empty selector, declared graph schemas, official initialization, separate readiness receipts and exact JSON storage representation without dropping identity or cardinality assertions.
  • Scoped Ruff formatting/checks, prose, whitespace and reach-census checks pass.
  • All required full hosted checks pass, including Cell acceptance. The earlier local source-tree copy check stopped at its memory cap and remains incomplete; it was not retried with relaxed limits. Earlier failed native runs remain recorded as failed evidence.
  • No local application stack, broad local suite or paid provider call ran.

AI assistance in the code (leave one; writing this description with AI is fine):

  • AI-generated: Codex prepared the workflow and checks, with independent source and actual-execution reviews.

  • I can explain every line of this diff and how it interacts with the rest of Loom.

Checklist

  • Required hosted checks pass. Local validation is limited to small checks related to this change.
  • The reach manifest records fixture call sites; no product schema or operation changes.
  • Application qualification is separate from package publication and marketplace acceptance.
  • Commits are signed off.

Exercise ordinary setup, discovery and scoped workflows with the candidate wheel and explicit owned cleanup on a disposable hosted runner.

Signed-off-by: zlareb1 <lareb.zafar@clickhouse.com>
Keep anonymous-volume refusal after mapping image storage to explicit project-owned volumes. Retain credentials only in memory and private files, and regenerate fixture operation reach.

Signed-off-by: zlareb1 <lareb.zafar@clickhouse.com>
Handle transport resets during readiness polling, keep permission refusals terminal, and record the failing startup stage without exposing response bodies.

Signed-off-by: zlareb1 <lareb.zafar@clickhouse.com>
The native tool workflows need the application and relational store, while the optional dataset subsystem stays disabled. Verify the resolved configuration and initialize only the fresh plugin storage bind under the pinned image user.

Signed-off-by: zlareb1 <lareb.zafar@clickhouse.com>
…ication

Signed-off-by: zlareb1 <lareb.zafar@clickhouse.com>
@vercel

vercel Bot commented Nov 21, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
loom-docs Ready Ready Preview Nov 21, 2026 1:39pm UTC

Request Review

The editor defaults endpoint omits start and end nodes. Construct their required fields directly and retain native tool execution checks. Fixed phase labels identify subsequent schema failures without recording response payloads.

Signed-off-by: zlareb1 <lareb.zafar@clickhouse.com>
The current Community Edition initializes both application and plugin storage before the daemon starts. Verify its fresh bind, completion dependency and ownership rather than running a second root helper. Record the two readiness observations separately.

Signed-off-by: zlareb1 <lareb.zafar@clickhouse.com>
The direct memory row query returns JSON text. Keep the exact identity tuple and row count, compare the encoded value, and retain observed rows before a mismatch fails the check.

Signed-off-by: zlareb1 <lareb.zafar@clickhouse.com>
The packaged setup guide must describe the native check and its candidate SDK scope while retaining the untested Cloud and version-matrix limitations.

Signed-off-by: zlareb1 <lareb.zafar@clickhouse.com>
@zlareb1
zlareb1 merged commit 0a00565 into main Nov 21, 2026
64 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants