Loading Historical Logs

API Endpoint

GET /api/executions/{executionId}/logs?page={page}&pageSize={pageSize}

// Response envelope
{
    "page"       : 1,
    "pageSize"   : 500,
    "totalCount" : 12400,
    "items"      : [ /* LogEntry[] */ ]
}

// LogEntry schema is identical to the SignalR payload
{
    "logId"     : "3f8e1a2c-...",
    "nodeId"    : "node-abc",
    "level"     : "Information",
    "message"   : "HTTP request completed",
    "timestamp" : "2026-05-21T14:33:01.123Z",
    "fields"    : { "executionId": "...", "tenantId": "..." },
    "traceId"   : "4bf92f3577b34da6..."
}

Client — executionLogsApiClient

// executionLogsApiClient.ts
export async function fetchExecutionLogs(
    executionId : string,
    page        : number,
    pageSize    : number = 500
): Promise<PagedResult<LogEntry>> {
    const response = await apiClient.get(
        `/api/executions/${executionId}/logs`,
        { params: { page, pageSize } }
    );
    return response.data;
}

Loading Strategy — Full Fetch with Pagination

For historical views the Observer Panel fetches all pages sequentially and calls appendLogs after each page. This loads the complete log set into executionStore so that in-memory filters (level, nodeId, text search) work over the full history without additional round-trips.

// useHistoricalLogs.ts
export async function loadAllHistoricalLogs(executionId: string) {
    const PAGE_SIZE = 500;
    let page = 1;
    let totalLoaded = 0;

    useExecutionStore.getState().setLoading(true);

    try {
        while (true) {
            const result = await fetchExecutionLogs(executionId, page, PAGE_SIZE);
            useExecutionStore.getState().appendLogs(result.items);
            totalLoaded += result.items.length;

            // appendLogs is idempotent — duplicates from live streaming are silently skipped

            if (totalLoaded >= result.totalCount) break;
            page++;
        }
    } finally {
        useExecutionStore.getState().setLoading(false);
    }
}

Invocation — When Logs Load

Deduplication During Mixed Load

When a live execution completes and the panel immediately loads historical logs, some entries may have already been appended via SignalR. The appendLogs action uses a Set-based logId check — duplicate entries are discarded without modifying state, so no re-render is triggered for already-known entries.

// executionStore.ts — appendLogs is always safe to call, even with overlapping data
appendLogs: (entries: LogEntry[]) => set(state => {
    const existingIds = new Set(state.logs.map(l => l.logId));
    const newEntries  = entries.filter(e => !existingIds.has(e.logId));

    if (newEntries.length === 0) return state;  // No change — no re-render

    return { logs: [...state.logs, ...newEntries] };
})

Backend — Log Storage

All log entries are written to Loki via the Serilog pipeline at emission time (see Backend Log Emission). The REST API reads from Loki using the executionId label as the query filter. Loki returns entries in insertion order; the API preserves this ordering in the paginated response.

// Loki label query used by the backend API handler
{executionId="3f8e1a2c-..."}
    | json
    | line_format "{{.message}}"

// Equivalent LogQL query for all entries in an execution
{executionId="3f8e1a2c-..."} | json | __error__=""