Error Logs β Admin CloudWatch Query Console
The SmartBot admin console includes an Error Logs panel that lets authorized users query aggregated error-level log entries across all SmartBot processing components for a specified time range. Use this panel to identify recurring error patterns, understand their frequency and distribution over time, and drill into individual error occurrences for full context.
Audience: SmartBot administrators responsible for monitoring conversation health and diagnosing errors across channels.
Before you begin
- You must have admin access to the SmartBot admin console. Contact your SmartBot administrator if you do not see the Error Logs panel.
- Error log queries run asynchronously. After you submit a query, the system returns a query ID immediately and continues processing in the background. You poll for results rather than waiting on a single request. Allow time for the query to complete, especially for longer time ranges.
- The Error Logs panel queries log data for all SmartBot processing components across the selected time range. Very broad time ranges (multiple weeks) may take longer to return results.
Steps
Submit an error log query
Sign in to the SmartBot admin console.
Navigate to the analytics section using the navigation panel.
Select Error Logs to open the error log query panel.
Set the time range for your query. You can use either:
- Relative time β specify a number of minutes, hours, days, or weeks before now (for example, "last 24 hours" or "last 7 days").
- Absolute window β specify an explicit start date/time and end date/time.
Submit the query.
Note: The system accepts your query and returns a query ID immediately. The query runs in the background β you do not need to keep the panel open while it processes.
When the query status changes to complete, the results load automatically in the panel.
Read error patterns in the results
Results are presented as error patterns β grouped summaries of similar log entries β rather than raw individual log lines. Each pattern row shows:
| Column | Description |
|---|---|
| Pattern | A representative string describing the class of error. Similar errors are grouped into a single pattern so you can see recurring issues at a glance. |
| Sample count | The number of log entries that matched this pattern within your selected time range. |
| Percentage | The share of total errors this pattern represents, expressed as a percentage. Use this to prioritize the most impactful error classes. |
| Time histogram | A visual distribution of when this pattern occurred across the query window. Spikes indicate bursts of errors at specific times. |
| Representative samples | Example log entries drawn from this pattern. Use these to understand what the error looks like in practice. |
Tip: Sort by percentage to surface the most frequent error classes first. A pattern affecting 40% of errors is a higher priority than one affecting 2%, even if the latter has a larger absolute count during an unusual spike.
Interpret error patterns to prioritize action
The error patterns view is designed to help you triage quickly:
- High-percentage patterns β Address these first. A single error class accounting for a large share of total errors typically points to a systemic issue such as a misconfigured center setting, an unavailable downstream service, or an invalid guest input triggering a consistent failure path.
- Histogram spikes β A pattern with a low overall count but a sharp spike at a specific time suggests a transient event (for example, a brief service outage or a high-volume conversation burst). Compare the spike time against any known deployments or center-level changes.
- Flat, low-percentage patterns β These may represent edge cases that occur infrequently. Review the representative samples to determine whether they require immediate action or can be monitored over time.
Worked example: Suppose your query for the last 7 days shows two dominant patterns. Pattern A β "Guest verification failed: OTP not delivered" β accounts for 38% of errors and shows a consistent daily occurrence. Pattern B β "Service lookup returned empty result" β accounts for 12% and shows a spike on Tuesday afternoon. You would investigate Pattern A first (likely an SMS delivery configuration issue for one or more centers) and separately investigate what changed on Tuesday for Pattern B (possibly a center's service catalog was temporarily unpublished).
Drill into a specific error occurrence
When a representative sample identifies a specific error you want to investigate further, use the surrounding logs drill-down to retrieve the full context for that occurrence.
- In the results panel, locate the representative sample that contains the error you want to investigate.
- Note the request ID and timestamp shown for that log entry.
- Select the drill-down action for that sample (the panel surfaces a surrounding logs option for each sample).
- The system retrieves the log lines immediately surrounding that request, giving you the complete sequence of events that led to and followed the error.
Note: The surrounding logs view shows activity scoped to the specific request ID and log group associated with that error. This lets you trace a single conversation error from start to finish without sorting through unrelated log entries.
Tips
- Use relative time ranges for routine monitoring. Querying "last 24 hours" or "last 48 hours" on a regular schedule helps you catch new error patterns before they affect a large number of guests.
- Use absolute time windows for incident investigation. When a guest or center reports a problem at a specific time, set an absolute window that brackets that period. This scopes results to the incident without noise from unrelated time periods.
- Compare results across time ranges to detect trends. Run the same pattern query for "last 7 days" and "last 30 days" and compare the percentage values. A pattern that was 2% last month but is 15% this week is growing and warrants investigation even if its absolute count is still small.
- Use the histogram to correlate errors with center events. If a pattern's histogram shows a spike aligned with a scheduled maintenance window or a high-traffic period (such as a weekend booking rush), the root cause is likely operational rather than a configuration defect.
- Drill into surrounding logs for any pattern where the representative sample alone does not explain the failure. The surrounding logs view provides the full request sequence and is the most efficient way to understand context without access to raw infrastructure tooling.
Troubleshooting
The results panel shows no patterns after the query completes.
The selected time range may contain no ERROR-level log entries, or there may have been no SmartBot activity during that window. Broaden the time range and resubmit. If the panel still returns no results across a multi-day window during which you know conversations occurred, contact your SmartBot administrator to verify that error log collection is active for your organization.
The query status does not update to complete.
Error log queries run asynchronously. For very broad time ranges (multiple weeks), queries may take several minutes to complete. Wait at least two to three minutes before concluding the query has stalled. If the status does not change after five minutes, close the panel and resubmit the query with a narrower time range to confirm the panel is functioning, then retry the broader range.
The surrounding logs drill-down returns no results for a sample.
The surrounding logs view requires a valid request ID and timestamp. If the representative sample does not include a request ID β for example, because the error occurred before a request ID was assigned β the drill-down has no anchor to retrieve context from. In this case, use the pattern string and timestamp to correlate with other error patterns in the same time window.
A pattern percentage changes significantly between two identical queries run minutes apart.
Error log query results reflect the state of the underlying log data at query submission time. If new errors arrive between two queries, the percentages recalculate based on the updated total. This is expected behavior for a live system. Use absolute time windows when you need consistent, reproducible results for a fixed period.
The error pattern string is not descriptive enough to identify the root cause.
Use the representative log samples in that pattern row. Samples contain more detail than the pattern string, which is a generalized representation of the error class. If the samples still do not provide enough context, use the surrounding logs drill-down on the most informative sample to retrieve the full request sequence.