fix(observability): fix top performers data loss, dead guards, display bugs, and response time unit conversion

[shoummu1]

Bug-fix PR

📌 Summary

Fixes five silent display/logic bugs in the Top Performers metrics panel (#3793) and resolves the response time unit inconsistency (seconds displayed as milliseconds) across all three monitoring panels in the Admin UI (#3580): Top Performers, KPI cards, and individual entity metrics cards.

🔗 Related Issues

Closes: #3793
Closes: #3580

🔁 Reproduction Steps

See #3793 for full steps. Minimal:

1. Navigate to Admin UI → Metrics → Top Performers
2. Observe "Last Used" column — raw ISO string (2026-03-13T06:00:00.000000) displayed
3. Observe "Success Rate" for a tool with no executions — shows 0.0% with red badge instead of N/A
4. Observe "Avg Response Time" in Top Performers and entity metrics cards — value shown in seconds (e.g. 1.194) instead of milliseconds (1194ms)
5. Register a tool with 0% success rate — the panel treats it as if it has no metrics at all

🐞 Root Cause

Six bugs across three files:

#	File	Bug
1	metrics_common.py	build_top_performers() used falsy checks (if result.success_rate, if result.avg_response_time) which coerced genuine 0.0 to None, losing the distinction between "no data" and actual zero
2	metrics_top_performers_partial.html	Dead != 'N/A' guard — avgResponseTime is Optional[float], never the string 'N/A'; branch was unreachable
3	metrics_top_performers_partial.html	Dead != 'Never' guard — lastExecution is None or an ISO string, never the string 'Never'; branch was unreachable
4	metrics_top_performers_partial.html	{{ last_exec }} rendered the raw ISO-8601 datetime verbatim instead of a human-readable format
5	metrics_top_performers_partial.html	item.get('successRate', 0) or 0 collapsed None (no executions) into 0, making un-executed tools display 0.0% with a red error badge
6	metrics_top_performers_partial.html, admin.js	avg_response_time (stored in seconds) was rendered without ×1000 conversion across Top Performers template, KPI cards, entity metrics cards, paginated table rows, and CSV export

💡 Fix Description

mcpgateway/utils/metrics_common.py

• Changed falsy checks to explicit is not None guards so 0.0 values are preserved:

  # Before
  avg_response_time=float(result.avg_response_time) if result.avg_response_time else None,
  success_rate=float(result.success_rate) if result.success_rate else None,
  # After
  avg_response_time=float(result.avg_response_time) if result.avg_response_time is not None else None,
  success_rate=float(result.success_rate) if result.success_rate is not None else None,

mcpgateway/templates/metrics_top_performers_partial.html

• Removed dead != 'N/A' guard; None check is sufficient
• Added ×1000 conversion for avgResponseTime (seconds → ms)
• Replaced item.get('successRate', 0) or 0 with a proper None-aware check; renders N/A (no badge) for un-executed tools
• Removed dead != 'Never' guard
• Replaced verbatim {{ last_exec }} with {{ last_exec[:16] | replace('T', ' ') }} → e.g. 2026-03-13 06:00

mcpgateway/static/admin.js

• extractKPIData: multiplied weighted average avgResponseTime by ×1000 before returning to KPI display
• createTopPerformersTable / updateTableRows: || → ?? (preserves 0.0) + ×1000 conversion applied consistently to both initial render and paginated updates
• exportMetricsToCSV: same ?? + ×1000 fix so downloaded CSV values match the UI
• createMetricsCard: added explicit avgResponseTime branch (×1000, .toFixed(1) ms) and lastExecutionTime branch (ISO slice → YYYY-MM-DD HH:mm)

🧪 Verification

Check	Command	Status
Lint suite	make lint	✅
Unit tests	make test	✅
Coverage ≥ 90 %	make coverage	✅
Manual regression no longer fails	Tested on UI	✅

Manual checks performed:

• Top Performers "Avg Response Time": 1194ms ✅
• Top Performers "Last Used": 2026-03-13 06:00 ✅
• Top Performers "Success Rate": 100.0% badge for executed tool ✅
• Resources/Prompts/Servers tabs: N/A (no badge) for un-executed entities ✅
• Tools Metrics card "Average Response Time": 1194.1 ms ✅
• Tools Metrics card "Last Execution Time": 2026-03-13 06:00 ✅

📐 MCP Compliance (if relevant)

• Matches current MCP spec
• No breaking change to MCP clients — display-only fixes, no API/schema changes

✅ Checklist

• Code formatted (make black isort pre-commit)
• No secrets/credentials committed

[shoummu1]

Hi @mxwlantian, Can you please help me review this PR? It includes changes for #3580

[crivetimihai]

Thanks @shoummu1 — good catch on the data loss and unit conversion issues. Small, targeted fix. Please add the DCO Signed-off-by line.

[crivetimihai]

Maintainer review notes

Rebased onto current main and reviewed the full changeset. The core fixes are correct — confirmed avg_response_time is stored in seconds throughout the stack (DB columns, rollup service, schemas all document "seconds"), so the ×1000 display conversion is right.

Rebase conflict resolution

mcpgateway/static/admin.js was split into modules under mcpgateway/admin_ui/ by #3137. All admin.js changes were transplanted to mcpgateway/admin_ui/metrics.js. Test conflicts in admin-parsing.test.js resolved by keeping main's structure with updated assertions.

Changes made during review

Bug fix — CSV export zero-value handling:
The exportMetricsToCSV function used a truthiness check on the ?? result, which would treat avg_response_time=0 as N/A. Changed to != null for consistency with the table rendering code.

Dead code removal — createPerformanceCard:
The PR added avgResponseTime/lastExecutionTime display branches to createPerformanceCard, but that function's metrics list only contains memoryUsage, cpuUsage, diskIo, networkThroughput, cacheHitRate, and activeThreads — those branches were unreachable. Reverted to the original simple display logic.

Trailing whitespace fix in updateTableRows.

Test fixes:

• Updated tests/unit/js/metrics.test.js assertion (150 → 150000) broken by the ×1000 conversion.
• Added 4 JS tests for createMetricsCard: seconds→ms conversion, ISO datetime formatting, null handling, zero-value preservation.
• Added Python test for build_top_performers verifying that 0.0 values for avg_response_time and success_rate are preserved (not coerced to None).

Verification

• All 89 metrics JS tests pass
• All 3 Python test_metrics_common tests pass
• Full JS suite: 2005 passed, 3 skipped