KQL Mastery

Try it yourself: All ✅ examples in this skill can be run against the public help cluster: https://help.kusto.windows.net, database Samples (contains StormEvents, SimpleGraph_Nodes/Edges, nyc_taxi, and more).

1. Running KQL with Fabric RTI MCP

Fabric RTI MCP exposes Kusto functionality as MCP tools. Authentication is handled transparently using Azure Identity.

Available tools

Query vs management commands

KQL has two execution planes, each with its own MCP tool:

Basic usage

# Query plane — use kusto_query
kusto_query(
    cluster_uri="https://help.kusto.windows.net",
    database="Samples",
    query="StormEvents | summarize count() by EventType | top 5 by count_ desc"
)

# Management plane — use kusto_command
kusto_command(
    cluster_uri="https://help.kusto.windows.net",
    database="Samples",
    command=".show tables"
)

# Schema exploration — use kusto_describe_database or kusto_describe_database_entity
kusto_describe_database(
    cluster_uri="https://help.kusto.windows.net",
    database="Samples"
)

# Sample data — use kusto_sample_entity
kusto_sample_entity(
    cluster_uri="https://help.kusto.windows.net",
    database="Samples",
    entity_name="StormEvents",
    entity_type="table",
    sample_size=5
)

# Graph queries — use kusto_graph_query
kusto_graph_query(
    cluster_uri="https://mycluster.kusto.windows.net",
    database="MyDB",
    graph_name="MyGraph",
    query="| graph-match (node) project labels=labels(node)"
)

# Deeplinks — use kusto_deeplink_from_query
kusto_deeplink_from_query(
    cluster_uri="https://help.kusto.windows.net",
    database="Samples",
    query="StormEvents | count"
)

Exploration workflow

When encountering a new cluster or database:

1. List entities: kusto_list_entities(cluster_uri, entity_type="tables", database="MyDB")
2. Get schema: kusto_describe_database_entity(entity_name="MyTable", entity_type="table", cluster_uri=..., database=...)
3. Sample data: kusto_sample_entity(entity_name="MyTable", entity_type="table", cluster_uri=..., sample_size=5)
4. Count rows: kusto_query(query="MyTable | count", cluster_uri=..., database=...)
5. Run analysis: kusto_query(query="MyTable | where ... | summarize ...", cluster_uri=..., database=...)

2. Dynamic Type Discipline

KQL's dynamic type is flexible but strict in certain contexts. A common mistake is using a dynamic column in summarize by, order by, or join on without casting.

The rule: Any time you use a dynamic-typed column in by, on, or order by, wrap it in an explicit cast.

// ❌ ERROR: "Summarize group key 'Partners' is of a 'dynamic' type"
| summarize count() by Partners

// ✅ FIX
| summarize count() by tostring(Partners)

// ❌ ERROR: "order operator: key can't be of dynamic type"
| order by Area desc

// ✅ FIX
| order by tostring(Area) desc

// ❌ ERROR in join: dynamic join key
| join kind=inner other on $left.Area == $right.Area

// ✅ FIX — cast both sides
| extend Area_str = tostring(Area)
| join kind=inner (other | extend Area_str = tostring(Area)) on Area_str

Self-correction: When you see "is of a 'dynamic' type" in an error, add tostring(), tolong(), or todouble().

3. Join Patterns & Pitfalls

KQL joins have constraints that differ from SQL.

Equality only

KQL join conditions support only `==`. No <, >, !=, or function calls in join predicates.

// ❌ ERROR: "Only equality is allowed in this context"
| join on geo_distance_2points(a.Lat, a.Lon, b.Lat, b.Lon) < 1000

// ✅ WORKAROUND — pre-bucket into spatial cells, then join on cell ID
| extend cell = geo_point_to_s2cell(Lon, Lat, 8)
| join kind=inner (other | extend cell = geo_point_to_s2cell(Lon, Lat, 8)) on cell

For range joins, pre-bin values: | extend bin_val = bin(Value, 100), then join on bin_val.

Left/right attribute matching

Both sides of a join on clause must reference column entities only — not expressions, not aggregates.

// ❌ ERROR: "for each left attribute, right attribute should be selected"
| join kind=inner other on $left.col1

// ✅ FIX — specify both sides explicitly
| join kind=inner other on $left.col1 == $right.col1

Cardinality check before large joins

Always check cardinality before joining tables with >10K rows. A cross-join explosion was the source of the single E_RUNAWAY_QUERY error (25K × 195 = potential 4.8M rows).

// Before joining, check how many rows each side contributes
TableA | summarize dcount(JoinKey)  // → 25,000? Too many for an unconstrained join
TableB | summarize dcount(JoinKey)  // → 195? OK if filtered first

4. Regex in KQL

KQL handles regex natively — no need for Python.

The extract_all gotcha

Unlike Python's re.findall(), KQL's extract_all requires capturing groups in the regex:

// ❌ ERROR: "extractall(): argument 2 must be a valid regex with [1..16] matching groups"
| extend words = extract_all(@"[a-zA-Z]{3,}", Text)

// ✅ FIX — add parentheses around the pattern
| extend words = extract_all(@"([a-zA-Z]{3,})", Text)

Regex toolkit — don't fall back to Python

5. Serialization Requirements

Window functions need serialized (ordered) input.

// ❌ ERROR: "Function 'row_cumsum' cannot be invoked. The row set must be serialized."
| summarize Online = sum(Direction) by bin(Timestamp, 5m)
| extend CumulativeOnline = row_cumsum(Online)

// ✅ FIX — add | serialize (or | order by, which implicitly serializes)
| summarize Online = sum(Direction) by bin(Timestamp, 5m)
| order by Timestamp asc
| extend CumulativeOnline = row_cumsum(Online)

Functions requiring serialization: row_number(), row_cumsum(), prev(), next(), row_window_session().

6. Memory-Safe Query Patterns

The most common memory error. Caused by scanning too much data without pre-filtering.

The progression of safety

Safest ──────────────────────────────────────────────── Most dangerous
| count    | take 10    | where + summarize    | summarize (no filter)    | full scan

Rules for large tables (>1M rows)

1. Always start with `| count` to understand table size
2. Always `| where` before `| summarize` — filter time range, partition key, or category first
3. Never `dcount()` on high-cardinality columns without pre-filtering
4. Check join cardinality before executing (see Section 3)
5. Use `materialize()` for subqueries referenced multiple times

// ❌ OUT OF MEMORY — 24M rows, no filter, dcount on every column
Consumption
| summarize dcount(Consumed), count() by Timestamp, HouseholdId, MeterType
| where dcount_Consumed > 1

// ✅ SAFE — filter first, then aggregate
Consumption
| where Timestamp between (datetime(2023-04-15) .. datetime(2023-04-16))
| summarize dcount(Consumed) by HouseholdId, MeterType
| where dcount_Consumed > 1

When you see E_LOW_MEMORY_CONDITION

The query touched too much data. Your options:

• Add | where filters (time range, partition key)
• Reduce the number of by columns in summarize
• Break into smaller time windows and union results
• Use | sample 10000 for exploratory work instead of full scans

When you see E_RUNAWAY_QUERY

A join or aggregation produced too many output rows. Check join cardinality — one or both sides is too large.

7. Result Size Discipline

Large results slow down analysis. Prevention:

The vector trap: Tables with embedding columns (1536-dim float arrays) produce ~30KB per row. Even | take 20 yields 600KB. Always | project away vector columns unless you specifically need them.

With MCP tools: Use kusto_sample_entity for quick data previews. For deeper exploration, use kusto_query with | take N or | top N to bound results.

8. String Comparison Strictness

KQL sometimes requires explicit casts when comparing computed string values — even when both sides are already strings.

// ❌ ERROR: "Cannot compare values of types string and string. Try adding explicit casts"
| where geo_point_to_s2cell(Lon, Lat, 16) == other_cell

// ✅ FIX — wrap both sides in tostring()
| where tostring(geo_point_to_s2cell(Lon, Lat, 16)) == tostring(other_cell)

This is most common with computed values from geo_point_to_s2cell() and strcat() comparisons. When in doubt, cast with tostring().

9. Advanced Functions

KQL handles these natively — no need for Python:

Vector similarity

// try it! — cosine similarity on Iris feature vectors
let target = pack_array(5.1, 3.5, 1.4, 0.2);
Iris
| extend Vec = pack_array(SepalLength, SepalWidth, PetalLength, PetalWidth)
| extend sim = series_cosine_similarity(Vec, target)
| top 5 by sim desc

Geo operations

// Distance between two points (meters)
StormEvents | extend dist = geo_distance_2points(BeginLon, BeginLat, EndLon, EndLat)

// Spatial bucketing for joins
StormEvents | extend cell = geo_point_to_s2cell(BeginLon, BeginLat, 8)

Graph queries

Use the kusto_graph_query MCP tool for graph traversal. It automatically handles graph snapshots when available:

// Persistent graph model — query the latest snapshot
graph("Simple")
| graph-match (src)-[e*1..5]->(dst)
  where src.name == "Alice"
  project src.name, dst.name, path_length = array_length(e)

// Transient graph — build inline with make-graph
SimpleGraph_Edges
| make-graph source --> target with SimpleGraph_Nodes on id
| graph-match (src)-[e*1..5]->(dst)
  where src.name == "Alice"
  project src.name, dst.name, path_length = array_length(e)

Time series

// Create a time series and detect anomalies
StormEvents
| make-series count() default=0 on StartTime step 1d
| extend anomalies = series_decompose_anomalies(count_)

For detailed examples and patterns, consult references/advanced-patterns.md.

10. Self-Correction Lookup Table

When you encounter an error, look it up here before retrying:

11. Datetime Pitfalls

Datetime literals are a common source of errors. A wrong literal format can cascade into completely different approaches instead of fixing the small issue.

Literal format

// ❌ WRONG — bare year is not a valid datetime
| where StartTime > datetime(2007)

// ✅ RIGHT — always use full date format
| where StartTime > datetime(2007-01-01)

Filtering by year, month, or hour

// ❌ WRONG — comparing datetime column to integer
| where StartTime == 2007

// ✅ RIGHT — use datetime_part() to extract components
| where datetime_part("year", StartTime) == 2007

// ✅ ALSO RIGHT — use between with datetime range
| where StartTime between (datetime(2007-01-01) .. datetime(2007-12-31T23:59:59))

Time bucketing in summarize

// This works, but can be harder to read and reuse in complex queries
| summarize count() by startofmonth(StartTime)

// Clearer — extend first, then summarize by the computed column
| extend Month = startofmonth(StartTime)
| summarize count() by Month
| order by Month asc

Useful datetime functions

12. Operator Naming & Equality

KQL has subtle differences from SQL syntax.

Naming conventions

Equality operators

// In where clauses, == is case-sensitive, =~ is case-insensitive
| where State == "TEXAS"      // exact match
| where State =~ "texas"      // case-insensitive
| where State != "TEXAS"      // not equal
| where State !~ "texas"      // case-insensitive not equal

// In joins, use == only
| join kind=inner other on $left.Key == $right.Key

sort vs order

Both sort by and order by work identically in KQL — they are aliases. Use whichever you prefer, but be consistent.

contains vs has

// contains: substring match (slower)
| where Message contains "error"        // finds "MyErrorHandler" too

// has: term/word match (faster, uses index)
| where Message has "error"             // matches word boundaries only

// For exact prefix/suffix
| where Message startswith "Error:"
| where Message endswith ".log"

13. Error Recovery Strategy

When a first KQL query fails, the temptation is to abandon the entire approach and try something completely different. The correct response is almost always to fix the specific error, not change strategy.

The pattern to avoid

Query 1: extract(@"pattern", 1, col)  → Parse error
Query 2: todynamic(col)               → Different error  
Query 3: parse_json(col)              → Another error
Query 4: Python script                → Works but 10x tokens

The correct pattern

Query 1: extract(@"pattern", 1, col)  → Parse error (bad escaping)
Query 2: extract(@"pattern", 1, col)  → Fix the specific escaping issue → Success

Rules for error recovery:

1. Read the error message carefully — it almost always tells you exactly what's wrong
2. Fix the specific syntax/escaping issue, don't switch approaches
3. Use the self-correction table (Section 10) to map errors to fixes
4. Only switch approaches after 2 failed fixes of the same query
5. The parse operator is often simpler than extract() for structured text:

// Instead of complex regex:
// extract(@"User '([^']+)' sent (\d+) bytes", 1, Message)

// Use parse for structured extraction:
| parse Message with * "User '" Username "' sent " ByteCount " bytes" *

14. Query Writing Checklist

Before running any KQL query, mentally check:

1. Pre-filtered? Large tables have a | where before any | summarize
2. Result bounded? Exploratory queries end with | take N or | top N
3. Dynamic columns cast? Any dynamic column in by/on/order by is wrapped
4. Regex has groups? extract_all patterns have () around what you want to capture
5. Join cardinality safe? Both sides checked with dcount() before joining
6. Needed columns only? Wide tables get | project to drop unneeded columns
7. Datetime literals valid? Using datetime(2024-01-01) not datetime(2024) or bare integers
8. Complex by-expressions? Use | extend first, then | summarize by the computed column
9. Error recovery plan? If a query fails, fix the specific error — don't change strategy
10. Right tool? Use kusto_query for queries, kusto_command for management commands, kusto_sample_entity for quick previews
11. Checked the plan? For expensive queries (joins, large tables), use kusto_show_queryplan to compare approaches before executing
12. Cluster healthy? Before heavy workloads, use kusto_diagnostics to check capacity and permissions

15. Diagnostics & Query Optimization

Two tools let you look before you leap: kusto_show_queryplan for query cost estimation and kusto_diagnostics for cluster health.

Query plan analysis

kusto_show_queryplan plans a query without executing it. Returns:

kusto_show_queryplan(
    query="Trips | where pickup_datetime > datetime(2014-01-01) | summarize count() by vendor_id",
    cluster_uri="https://help.kusto.windows.net",
    database="Samples"
)

What queryplan CAN detect

• Expensive joins: Self-joins double estimated_rows and shard_scans count.
• Materialize overhead: materialize() + join has higher PlanSize than single-pass multi-aggregation.
• Precomputed results: Bare | count returns estimated_rows=1 and ConstantDataTable in the tree — no scan.
• Column/table typos: Returns an error field with the semantic error message.
• Filter presence: has_selection=true vs false shows whether a where clause narrows the scan.

What queryplan CANNOT detect

• Data volume differences: Full scan vs filtered scan on the same table show similar estimated_rows (both report table size). Use has_selection instead — false means full scan.
• Type mismatches: KQL auto-casts, so where fare_amount == "expensive" plans successfully (returns 0 rows at runtime).
• Ambiguous join columns: KQL resolves these at plan time without error.

Comparing two query approaches

The pattern: plan both, compare estimated_rows and shard_scans.

# Plan A: direct summarize
plan_a = kusto_show_queryplan(query="Trips | summarize count() by vendor_id, payment_type", ...)

# Plan B: self-join (looks "clever" but is worse)
plan_b = kusto_show_queryplan(query="Trips | as T | join kind=inner (T | summarize by vendor_id) on vendor_id | summarize count() by payment_type", ...)

# Compare: plan_b.execution_hints.estimated_rows will be 2x plan_a's → pick A

Regression thresholds: Flag a rewrite as a regression if estimated_rows increases >50% or shard_scans count increases >30%.

Cluster diagnostics

kusto_diagnostics runs 7 commands (best-effort — permission failures don't block others):

kusto_diagnostics(
    cluster_uri="https://help.kusto.windows.net",
    database="Samples"
)

When to use diagnostics

• Before heavy workloads: Check capacity — if Queries.Remaining is low, wait or batch.
• Batching decisions: batch_size = min(remaining_slots / 2, num_tasks) — leave 50% headroom.
• Permission checks: principal_roles tells you what you can do before you try and fail.
• Debugging ingestion issues: ingestion_failures surfaces errors from the last 24h.