FIX: Reduce logs panel noise and make execution messages readable (OR-1457)

[abhizipstack]

What

• Demoted `UsingCachedObject` from `InfoLevel` to `DebugLevel` so the per-query "Using Cached Object of bigquery_connection_object" spam no longer reaches the UI at default verbosity.
• Fixed `Processing Model : ""` — use `cls.name` (fallback `file_name`) instead of `str(cls)` which was empty for dynamic no-code model classes.
• Stopped `str(Enum)` from leaking dotted enum names (`ExecStatus.Success`, `Materialization.EPHEMERAL`, etc.) into user-visible messages; use `.value` so the UI shows `SUCCESS` / `OK` / `TABLE`.
• Fixed the summary counter bug — `parse_and_fire_reports` compared `end_status` against `ExecStatus.Success`/`Error`, but `end_status` is set to `OK`/`Fail`. Counters stayed at zero, so `DONE PASS=0 WARN=0 ERROR=0 SKIP=0 TOTAL=0` showed even after successful runs. Now compares against correct end-status values.
• Added a log-level selector to the bottom logs tab: `All logs / Info & above / Warnings & errors / Errors only`. Default `Info & above` (hides debug noise out of the box). Choice persists in `localStorage`.
• Each log row is tinted by severity (grey / default / amber / red) for quick scanning.

Why

Bottom logs panel was dominated by dev-internal debug lines and leaky enum reprs:
```
06:16:09 [info] [ThreadPool]: Processing Model : "".
06:16:09 [info] [ThreadPool]: Executing Model Node: ... Materialization: Materialization.EPHEMERAL ...
06:16:09 [info] [ThreadPool]: 1 of 2 ExecStatus.Success .................................... [ ExecStatus.OK ]
06:16:09 [info] [ThreadPool]: DONE PASS=0 WARN=0 ERROR=0 SKIP=0 TOTAL=0
```
Users couldn't tell what actually happened after an action. After this change, the same run reads:
```
06:16:09 [info] [ThreadPool]: Processing Model : "MModelA".
06:16:09 [info] [ThreadPool]: Executing Model Node: ... Materialization: TABLE ...
06:16:09 [info] [ThreadPool]: 1 of 2 SUCCESS MModelA .......... [ OK ]
06:16:09 [info] [ThreadPool]: DONE PASS=2 WARN=0 ERROR=0 SKIP=0 TOTAL=2
```

How

Backend (`backend/visitran/`)

• `events/types.py:472` — `UsingCachedObject(InfoLevel) → DebugLevel`.
• `events/printer.py:89-96` — compare `end_status` to `OK`/`Fail`/`Warn`/`Skipped` values (was comparing to `Success`/`Error`).
• `visitran.py` — replaced `str(ExecStatus.X)` with `ExecStatus.X.value` everywhere (5 call sites). Same for `node.materialization`.
• `visitran.py:515` — `ProcessingModel(cls=getattr(cls, "name", "") or file_name)`.

Frontend (`frontend/src/ide/editor/no-code-model/no-code-model.jsx`)

• Socket handler stores `{ level, message }` (was a raw string) and reads `data?.data?.level`.
• New `Select` above the logs tab bound to `logsLevel` state, initialized from `localStorage.getItem("visitran.logsLevel") || "info"`.
• `LOG_LEVEL_RANK` threshold filters the rendered list; `LOG_LEVEL_COLOR` tints rows.

Can this PR break any existing features?

Low risk:

• Log socket payload already carried `level` — we just started using it. Missing level falls back to `info`.
• Old string-only entries in `logsInfo` would have crashed the render, but the list starts empty per session and no other writer exists.
• Enum `.value` change: external consumers that parsed `BaseResult.status == "ExecStatus.Success"` would break, but the only consumer is `parse_and_fire_reports` in the same commit and it's been updated.
• `UsingCachedObject` demotion to debug means operators who rely on seeing cached-connection hits must select "All logs" (documented via the dropdown label).

Database Migrations

None.

Env Config

None.

Relevant Docs

None.

Related Issues or PRs

• OR-1457

Dependencies Versions

No changes.

Notes on Testing

1. Open a no-code model backed by BigQuery, run it. Logs panel should no longer flood with "Using Cached Object of bigquery_connection_object".
2. Add a transformation (e.g. Sort) to a Postgres-backed model; the "Processing Model" line should show the actual model name (not empty quotes), and "Executing Model Node" should show `Materialization: TABLE` not `Materialization.TABLE`.
3. After a successful run, the summary line should read `DONE PASS=N ...` with the correct non-zero count.
4. Change the log-level selector to "All logs" and reload the page — selection should persist.
5. Trigger an error; the row should appear in red when the selector is "Errors only".

Screenshots

No screenshots available — the log-level filter and severity-tinted rows require a WebSocket connection for live log streaming, which is not available in the React dev server proxy setup. The feature works in production where nginx handles WebSocket forwarding.

UI changes:

• Log-level selector dropdown in the bottom Logs tab: All logs / Info & above / Warnings & errors / Errors only (persisted in localStorage)
• Each log row is tinted by severity: grey for debug, default for info, amber for warnings, red for errors

Checklist

TBD

Checklist

I have read and understood the Contribution Guidelines.

[greptile-apps]

Greptile Summary

This PR fixes several user-visible issues in the logs panel: enum repr leakage (ExecStatus.Success → SUCCESS), the empty Processing Model : "" string, broken summary counters (PASS/ERROR now increment correctly), and per-query cache-object spam. A log-level filter with localStorage persistence and per-row severity tinting are added on the frontend.

• warn_count and skip_count in parse_and_fire_reports are now correctly compared (.value instead of str()), but no code path in visitran.py sets end_status to "WARN" or "SKIPPED", so those two counters will still always be 0 in the DONE summary.

Confidence Score: 5/5

Safe to merge — all fixes are correct and the single finding is a pre-existing limitation, not a regression introduced by this PR.

All P1 bugs targeted by the PR are correctly fixed. The only comment is a P2 noting that warn/skip counters remain 0 because no code paths produce those end_status values, which is a pre-existing gap rather than a new regression. Frontend changes are well-structured with proper memoization and socket cleanup.

No files require special attention.

Important Files Changed

Filename	Overview
backend/visitran/events/printer.py	Fixed counter comparisons to use .value and elif chains; logic is correct, though no code path in visitran.py currently produces a "WARN" or "SKIPPED" end_status so those counters remain 0.
backend/visitran/events/types.py	Single-line demotion of UsingCachedObject from InfoLevel to DebugLevel to suppress per-query cache spam from the UI. Change is correct and intentional.
backend/visitran/visitran.py	Replaced all str(ExecStatus.X) usages with ExecStatus.X.value and fixed ProcessingModel class name lookup; all six call sites updated consistently.
frontend/src/ide/editor/no-code-model/no-code-model.jsx	Added log-level Select filter with localStorage persistence, per-row severity tinting, useMemo for filteredLogs, and improved socket cleanup with explicit disconnect. Changes are well-structured.

Sequence Diagram

sequenceDiagram
    participant BE as Backend (visitran.py)
    participant EV as events/printer.py
    participant SK as WebSocket
    participant FE as NoCodeModel (JSX)

    BE->>BE: run node (success/fail)
    BE->>EV: BaseResult(end_status=ExecStatus.OK.value / Fail.value)
    BE->>EV: parse_and_fire_reports()
    EV->>EV: compare end_status against OK/Fail/Warn/Skipped .value
    EV->>SK: fire_event(SummaryReport, EndSummaryReportCounts)

    SK-->>FE: logs:{socketSessionId} { level, message }
    FE->>FE: setLogsInfo([...old, { level, message }])
    FE->>FE: filteredLogs = useMemo(filter by LOG_LEVEL_RANK[logsLevel])
    FE->>FE: render rows tinted by LOG_LEVEL_COLOR[el.level]

    note over FE: logsLevel persisted in localStorage (default: info)

Loading

Prompt To Fix All With AI

This is a comment left during a code review.
Path: backend/visitran/events/printer.py
Line: 91-96

Comment:
**`warn_count` and `skip_count` will still always be zero**

The comparisons against `ExecStatus.Warn.value` ("WARN") and `ExecStatus.Skipped.value` ("SKIPPED") are now logically correct, but a grep of `visitran.py` shows that no code path ever sets `end_status` to either of those values — only `ExecStatus.OK.value` and `ExecStatus.Fail.value` are assigned. As a result, the DONE line will still always show `WARN=0 SKIP=0` regardless of what actually happened. This is a pre-existing gap that the counter fix doesn't fully close; if warn/skip paths are planned, the corresponding `end_status` assignments in `visitran.py` will need to be added.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (3): Last reviewed commit: "fix: use antd theme token for log-panel ..." | Re-trigger Greptile}

[abhizipstack]

Socket cleanup fix (d281ec8): The socket useEffect cleanup now properly calls newSocket.disconnect() to tear down the connection on unmount. Additionally, the listener is registered and unregistered using a closure-captured socketSessionId variable (assigned when the server sends session_id), ensuring the cleanup always references the correct session ID rather than the outer-scope cookie value.

[wicky-zipstack]

LGTM overall

[Deepak-Gnanasekar]

Review — APPROVED

Reviewed all 4 changed files (+94 / -43). Clean, targeted fixes with no new logic issues introduced.

Backend

printer.py — The root cause of PASS=0 WARN=0 ERROR=0 SKIP=0 TOTAL=0 was comparing end_status against ExecStatus.Success/ExecStatus.Error when it's actually set to OK/Fail. Fix is correct. The if → elif tightening is also appropriate since end_status values are mutually exclusive.

types.py — UsingCachedObject demotion from InfoLevel to DebugLevel is a clean one-liner. Users who need it can switch to "All logs" via the new dropdown.

visitran.py — All str(ExecStatus.X) → ExecStatus.X.value replacements are consistent across all 5 call sites (models, tests, seeds, snapshots). The hasattr guard on node.materialization.value is a good defensive choice. getattr(cls, "__name__", "") or file_name correctly handles dynamic no-code model classes that lacked a meaningful __str__.

Frontend (no-code-model.jsx)

• LOG_LEVEL_RANK / LOG_LEVEL_COLOR are correctly defined outside the component (no re-creation on render).
• filteredLogs wrapped in useMemo with correct [logsInfo, logsLevel] deps.
• localStorage persistence for log level is straightforward.
• Socket cleanup fix is a genuine improvement: capturing socketSessionId via closure ensures cleanup targets the right listener, and newSocket.disconnect() properly tears down the connection on unmount. The old code referenced the outer-scope sessionId (cookie value) which could diverge from the server-assigned session.
• dangerouslySetInnerHTML usage is pre-existing and parseLog sanitizes via DOMPurify — no new XSS surface.

Minor note (non-blocking)

• If a log message arrives before the session_id event (unlikely but theoretically possible with race conditions), it would be silently dropped. Current behavior is fine since the socket listener is only registered after session_id fires anyway.

LGTM — safe to merge.

[tahierhussain]

@abhizipstack Can you please add screenshots for the UI changes addressed in this PR?

[tahierhussain]

LGTM