Time-Series Analysis with timechart

The timechart verb bins data over time and applies aggregation functions to create time-series visualizations. This skill teaches you how to analyze trends and patterns in your data over specific time periods using OPAL.

When to Use This Skill

• Visualizing trends over time (error rate by hour, request volume trends)
• Creating time-series charts for dashboards
• Tracking how metrics change throughout the day/week/month
• Comparing behavior across time periods
• Identifying spikes or anomalies in temporal data

Note: timechart returns multiple rows per group (one per time bin) for time-series visualization. For single summary rows, see the aggregating-event-datasets skill.

Prerequisites

• Access to Observe tenant via MCP
• Understanding of event or interval datasets
• Familiarity with aggregation concepts (see aggregating-event-datasets skill)

Key Concepts

timechart - Temporal Binning

timechart groups data into time buckets (bins) and aggregates within each bucket:

• Returns multiple rows per group (one row per time bin)
• Adds temporal columns: _c_valid_from, _c_valid_to, _c_bucket
• Default: 300 bins (Observe picks optimal size)
• Can specify bin duration: 1h, 5m, 1d, 30s, etc.
• Can control bin count: options(bins: N)

Syntax:

timechart [bin_duration], aggregation_function(), group_by(dimension1, dimension2, ...)

timechart vs statsby

Example comparison:

statsby (summary):

statsby count(), group_by(namespace)
# Output: 1 row per namespace

timechart (time-series):

timechart 1h, count(), group_by(namespace)
# Output: 24 rows per namespace (for 24h query)

Temporal Output Columns

timechart adds three columns to output:

• _c_valid_from - Start of time bin (nanosecond timestamp)
• _c_valid_to - End of time bin (nanosecond timestamp)
• _c_bucket - Bucket identifier (integer)

Discovery Workflow

Same as other event/interval skills:

Step 1: Find dataset

discover_context("kubernetes logs")

Step 2: Get schema

discover_context(dataset_id="YOUR_DATASET_ID")

Note fields for filtering, grouping, and aggregating.

Basic Patterns

Pattern 1: Fixed Bin Duration

Use case: Hourly error count over 24 hours

filter contains(body, "error")
| timechart 1h, count()

Explanation: Bins data into 1-hour buckets, counts errors in each bucket. For 24h time range, returns 24 rows.

Output:

_c_valid_from,_c_valid_to,count,_c_bucket
1763110800000000000,1763114400000000000,4,489753
1763107200000000000,1763110800000000000,5,489752
...

Pattern 2: Time-Series by Dimension

Use case: Error count per hour, grouped by namespace

filter contains(body, "error")
| make_col namespace:string(resource_attributes."k8s.namespace.name")
| timechart 1h, count(), group_by(namespace)

Explanation: Returns multiple rows per namespace - one row for each hour. Suitable for multi-line chart showing errors over time per namespace.

Output:

_c_valid_from,_c_valid_to,namespace,count,_c_bucket
1763110800000000000,1763114400000000000,kube-system,4,489753
1763107200000000000,1763110800000000000,kube-system,5,489752
1763100000000000000,1763103600000000000,observe,16,489750
1763100000000000000,1763103600000000000,kube-system,5,489750
...

Pattern 3: Controlling Bin Count

Use case: Get exactly 10 data points across time range

filter contains(body, "error")
| make_col namespace:string(resource_attributes."k8s.namespace.name")
| timechart options(bins: 10), count(), group_by(namespace)

Explanation: Observe automatically calculates bin duration to produce at most 10 bins across the query time range.

Pattern 4: Multiple Aggregations

Use case: Track multiple metrics over time

make_col namespace:string(resource_attributes."k8s.namespace.name")
| timechart 1h,
    total_logs:count(),
    group_by(namespace)

Explanation: Calculate multiple aggregations per time bin. Each metric becomes a column in the output.

Pattern 5: Default Auto-Binning

Use case: Let Observe pick optimal bin size

filter contains(body, "error")
| make_col namespace:string(resource_attributes."k8s.namespace.name")
| timechart count(), group_by(namespace)

Explanation: Without specifying duration or bins, Observe defaults to 300 bins with optimal size for the time range.

Complete Example

End-to-end workflow for creating an error rate dashboard.

Scenario: Create a time-series chart showing error trends per namespace over the last 24 hours.

Step 1: Discovery

discover_context("kubernetes logs")

Found: Dataset "Kubernetes Explorer/Kubernetes Logs" (ID: 42161740)

Step 2: Build query

filter contains(body, "error") or contains(body, "ERROR")
| make_col namespace:string(resource_attributes."k8s.namespace.name")
| timechart 1h, error_count:count(), group_by(namespace)

Step 3: Execute

execute_opal_query(
    query="[query above]",
    primary_dataset_id="42161740",
    time_range="24h"
)

Step 4: Interpret results

_c_valid_from,_c_valid_to,namespace,error_count,_c_bucket
1763110800000000000,1763114400000000000,kube-system,4,489753
1763107200000000000,1763110800000000000,kube-system,5,489752
1763103600000000000,1763107200000000000,kube-system,5,489751
1763100000000000000,1763103600000000000,observe,16,489750
1763100000000000000,1763103600000000000,kube-system,5,489750
...

Analysis:

• Each row represents one hour for one namespace
• kube-system has relatively stable error rate (4-7 per hour)
• observe namespace shows a spike (16 errors in one hour)
• Data suitable for line chart with multiple series (one per namespace)

Visualization:

• X-axis: Time (_c_valid_from or _c_valid_to)
• Y-axis: error_count
• Series/Lines: One per namespace

Advanced Patterns

Pattern 6: Bin Duration Options

Common bin durations:

# Fine-grained (short time ranges)
timechart 30s, count()   # 30 second bins
timechart 1m, count()    # 1 minute bins
timechart 5m, count()    # 5 minute bins

# Medium (hours to days)
timechart 1h, count()    # 1 hour bins
timechart 6h, count()    # 6 hour bins

# Coarse (days to weeks)
timechart 1d, count()    # 1 day bins
timechart 1w, count()    # 1 week bins

Choosing bin size:

• Match your query time range: 1h range → 1m or 5m bins, 7d range → 1h or 6h bins
• Consider visualization: Too many bins = cluttered chart, too few = lost detail
• Start with auto-binning, adjust as needed

Pattern 7: Rate Calculations

Use case: Calculate error rate (errors per second)

filter contains(body, "error")
| make_col namespace:string(resource_attributes."k8s.namespace.name")
| timechart 1h, error_count:count(), group_by(namespace)
| make_col error_rate:float64(error_count)/3600.0  # errors per second (1h = 3600s)

Explanation: Adds derived column calculating rate per second from hourly counts.

Common Pitfalls

Pitfall 1: Using statsby When You Want Time-Series

❌ Wrong (if you want trends):

statsby count(), group_by(namespace)
# Returns 1 row per namespace (no time dimension)

✅ Correct:

timechart 1h, count(), group_by(namespace)
# Returns multiple rows per namespace (time-series)

Why: statsby gives totals, timechart gives trends over time.

Pitfall 2: Too Many Bins

❌ Wrong:

timechart 1s, count()  # For 24h range = 86,400 bins!

✅ Correct:

timechart 1m, count()  # For 24h range = 1,440 bins (manageable)
# OR
timechart options(bins: 100), count()  # Let Observe pick duration

Why: Too many bins overwhelm visualization and query performance.

Pitfall 3: Confusing Timestamps

❌ Wrong:

# Trying to use timestamp field directly for binning
filter timestamp > 1763100000000000000
| statsby count(), group_by(timestamp)

✅ Correct:

# Use timechart for temporal binning
timechart 1h, count()

Why: timechart automatically handles time binning - don't need manual timestamp grouping.

Pitfall 4: Expecting Single Summary

❌ Wrong (if you want total):

timechart 1h, count()
# Returns 24 rows for 24h range - need to sum them manually!

✅ Correct (for total):

statsby count()
# Returns 1 row with total

Why: timechart is for time-series, not totals. Use statsby for summaries.

Tips and Best Practices

• Start with defaults: Use timechart count() first, adjust bin size after seeing results
• Match bin to time range: Short ranges (1h) → small bins (1m), long ranges (30d) → large bins (1d)
• Name your metrics: Use error_count:count() not just count()
• Filter first: Apply filters before timechart for better performance
• Visualization-ready: Output is designed for time-series charts
• Test with small ranges: Start with 1h or 24h, expand once query works
• options(bins: N): Use when you want specific number of data points

Bin Duration Reference

Time Units:

• s - seconds
• m - minutes
• h - hours
• d - days
• w - weeks

Examples:

• 30s - 30 seconds
• 5m - 5 minutes
• 1h - 1 hour
• 6h - 6 hours
• 1d - 1 day
• 1w - 1 week

Default Behavior:

• No duration specified: options(bins: 300) - Observe picks optimal size
• options(bins: 1) - Single bin (equivalent to statsby)
• options(bins: N) - At most N bins across time range

Additional Resources

For more details, see:

• RESEARCH.md - Tested patterns including timechart vs statsby comparison
• OPAL Documentation - Official OPAL docs

Related Skills

• [aggregating-event-datasets] - For single summary rows (statsby)
• [filtering-event-datasets] - For filtering before time-series aggregation
• [analyzing-interval-datasets] - timechart works on Intervals too (spans, resources)

Last Updated: November 14, 2025 Version: 1.0 Tested With: Observe OPAL v2.x