SKILL.md
$28
DQL Reference Index
Use this index to route from a function group (e.g. time functions, conversions) to its detailed spec, or from a function name to its spec file.
Description
Items
array, binary, boolean, double, duration, long, record, string, timeframe, timestamp, uid
bucket, dataObject, dplPattern, entityAttribute, entitySelector, entityType, enum, executionBlock, expressionTimeseriesAggregation, expressionWithConstantValue, expressionWithFieldAccess, fieldPattern, filePattern, identifierForAnyField, identifierForEdgeType, identifierForFieldOnRootLevel, identifierForNodeType, joinCondition, jsonPath, metricKey, metricTimeseriesAggregation, namelessDplPattern, nonEmptyExecutionBlock, prefix, primitiveValue, simpleIdentifier, tabularFileExisting, tabularFileNew, url
append, data, dedup, describe, expand, fetch, fields, fieldsAdd, fieldsFlatten, fieldsKeep, fieldsRemove, fieldsRename, fieldsSnapshot, fieldsSummary, filter, filterOut, join, joinNested, limit, load, lookup, makeTimeseries, metrics, parse, search, smartscapeEdges, smartscapeNodes, sort, summarize, timeseries, traverse
avg, collectArray, collectDistinct, correlation, count, countDistinct, countDistinctApprox, countDistinctExact, countIf, max, median, min, percentRank, percentile, percentileFromSamples, percentiles, stddev, sum, takeAny, takeFirst, takeLast, takeMax, takeMin, variance
arrayAvg, arrayConcat, arrayCumulativeSum, arrayDelta, arrayDiff, arrayDistinct, arrayFirst, arrayFlatten, arrayIndexOf, arrayLast, arrayLastIndexOf, arrayMax, arrayMedian, arrayMin, arrayMovingAvg, arrayMovingMax, arrayMovingMin, arrayMovingSum, arrayPercentile, arrayRemoveNulls, arrayReverse, arraySize, arraySlice, arraySort, arraySum, arrayToString, vectorCosineDistance, vectorInnerProductDistance, vectorL1Distance, vectorL2Distance
bitwiseAnd, bitwiseCountOnes, bitwiseNot, bitwiseOr, bitwiseShiftLeft, bitwiseShiftRight, bitwiseXor
exists, in, isFalseOrNull, isNotNull, isNull, isTrueOrNull, isUid128, isUid64, isUuid
asArray, asBinary, asBoolean, asDouble, asDuration, asIp, asLong, asNumber, asRecord, asSmartscapeId, asString, asTimeframe, asTimestamp, asUid
e, pi
toArray, toBoolean, toDouble, toDuration, toIp, toLong, toSmartscapeId, toString, toTimeframe, toTimestamp, toUid, toVariant
array, duration, ip, record, smartscapeId, timeframe, timestamp, timestampFromUnixMillis, timestampFromUnixNanos, timestampFromUnixSeconds, uid128, uid64, uuid
hashCrc32, hashMd5, hashSha1, hashSha256, hashSha512, hashXxHash32, hashXxHash64
classicEntitySelector, entityAttr, entityName
Functions — Time series aggregation for expressions
avg, count, countDistinct, countDistinctApprox, countDistinctExact, countIf, end, max, median, min, percentRank, percentile, percentileFromSamples, start, sum
coalesce, if
jsonField, jsonPath, lookup, parse, parseAll, type
arrayElement, getEnd, getHighBits, getLowBits, getStart
iAny, iCollectArray, iIndex
abs, acos, asin, atan, atan2, bin, cbrt, ceil, cos, cosh, degreeToRadian, exp, floor, hexStringToNumber, hypotenuse, log, log10, log1p, numberToHexString, power, radianToDegree, random, range, round, signum, sin, sinh, sqrt, tan, tanh
ipIn, ipIsLinkLocal, ipIsLoopback, ipIsPrivate, ipIsPublic, ipMask, isIp, isIpV4, isIpV6
getNodeField, getNodeName
concat, contains, decodeBase16ToBinary, decodeBase16ToString, decodeBase64ToBinary, decodeBase64ToString, decodeUrl, encodeBase16, encodeBase64, encodeUrl, endsWith, escape, getCharacter, indexOf, lastIndexOf, levenshteinDistance, like, lower, matchesPattern, matchesPhrase, matchesRegex, matchesValue, punctuation, replacePattern, replaceString, splitByPattern, splitString, startsWith, stringLength, substring, trim, unescape, unescapeHtml, upper
formatTimestamp, getDayOfMonth, getDayOfWeek, getDayOfYear, getHour, getMinute, getMonth, getSecond, getWeekOfYear, getYear, now, unixMillisFromTimestamp, unixNanosFromTimestamp, unixSecondsFromTimestamp
Functions — Time series aggregation for metrics
avg, count, countDistinct, end, max, median, min, percentRank, percentile, start, sum
Syntax Pitfalls
❌ Wrong
✅ Right
Issue
filter field in ["a", "b"]
filter in(field, {"a", "b"})
[ and ] wrap sub-queries in DQL but do not wrap static array literals. Use {} or array() for static values.
filter: { in(field, [sub-query]) } (e.g. in timeseries filter:)
filter: { field in [sub-query] }
in() does not accept execution blocks as arguments. When the right-hand side is a sub-query (execution block), use the in operator: field in [execution block].
by: severity, status
by: {severity, status}
List of fields must be grouped by curly braces in by: clauses (summarize, makeTimeseries, etc.).
contains(toLowercase(field), "err")
contains(field, "err", false)
Don't wrap in lower() for case-insensitive matching. contains() has a built-in third positional caseSensitive parameter (default true).
filter name == "*serv*9*"
filter matchesValue(name, "*serv*") and matchesValue(name, "*9*")
== does not support wildcards. matchesValue() supports * wildcards but only at the beginning and/or end of the pattern—split mid-string wildcard intent into multiple calls combined with and.
matchesValue(field, "prod") on string field
contains(field, "prod")
Without wildcards, matchesValue() performs an exact (case-insensitive) match — it will not find "production". Use contains() for substring matching (or matchesValue(field, "*prod*") for wildcard matching).
toLowercase(field)
lower(field)
The function is lower(), not toLowercase(). Only type-casting functions use the to prefix (toString(), toLong(), etc.).
arrayAvg(field[]) or arraySum(field[])
arrayAvg(field) or field[]
field[] = element-wise iterative expression (array→array); arrayAvg(field) = collapse to scalar (array→single value). Never mix both — arrayAvg(field[]) is semantically wrong.
my_field after lookup or join
lookup.my_field / right.my_field
lookup prefixes added fields with lookup. by default (configurable via prefix:). join prefixes right-side fields with right..
substring(field, 0, 200)
substring(field, from: 0, to: 200)
The first parameter (expression) is positional, but from: and to: are named optional parameters and must include their names.
filter host = "A"
filter host == "A"
DQL uses == for equality comparison, not =. Single = is assignment (e.g., in fieldsAdd, summarize aliases).
fetch logs, from: toTimestamp('2026-01-01')
fetch logs, from: -24h
from: / to: accept duration literals (e.g., -24h, -7d) or now() expressions — not toTimestamp(). For absolute ranges use timeframe: "start/end" (ISO 8601).
filter log.level == "ERROR"
filter loglevel == "ERROR"
Log severity field is loglevel (no dot) — log.level does not exist.
sort count() desc
sort count() desc
Fields with special characters (like parentheses) must be wrapped in backticks.
length(field)
stringLength(field)
DQL string length function is stringLength — there is no length().
metrics dt.host.cpu.usage
timeseries avg(dt.host.cpu.usage)
metrics loads metric metadata, not values — use timeseries for data.
join [...], on:{left.a.b == right.a.b}
join [...], on:{left[a.b] == right[a.b]}
Dotted field names in join/lookup conditions require bracket notation with backticks.
fieldsSummary (no arguments)
fieldsSummary field1, field2
fieldsSummary requires at least one field parameter.
timeseries with percentile/median/percentRank — no results
Add rollup: avg (or min/max/sum) to the timeseries command
These three functions **require rollup:** on gauge/count metrics — without it the query silently returns empty.
lookup [...], fields: {dotted.name}
lookup [...], fields: {dotted.name}
Do not backtick field names inside the fields: parameter of lookup — causes PARSE_ERROR.
data record(key: "val")
data record(key = "val")
record() uses = for named fields, not : — : is for command parameters like rollup:.
getNodeField(dt.smartscape.host, "tags")["tag.key"]
getNodeField(dt.smartscape.host, "tags")[tag.key]
In this tag-map access pattern, bracket keys must use unquoted identifier syntax; quoted keys cause a parse error.
by: {dt.entity.host} or dt.entity.*
by: {dt.smartscape.host} or dt.smartscape.*
dt.entity.* is deprecated — always use dt.smartscape.* in new queries.
Fetch Command → Data Model
DQL queries start with fetch <data_object> or timeseries. There is **no fetch dt.metric** — metrics use timeseries.
Fetch Command
Data Model
Key Fields / Notes
fetch spans
Distributed tracing
span.*, service.*, http.*, db.*, code.*, exception.*
fetch logs
Log events
log.*, k8s.*, host.* — message body is content, severity is loglevel (NOT log.level)
fetch events
DAVIS / infra events
event.*, dt.smartscape.*
fetch bizevents
Business events
event.*, custom fields
fetch security.events
Security events
vulnerability.*, event.*
fetch user.sessions
RUM sessions
dt.rum.*, browser.*, geo.*
fetch user.events
RUM individual events
page views, clicks, requests, errors
fetch user.replays
Session replay recordings
fetch application.snapshots
Application snapshots
fetch dt.davis.events
Davis-detected events
fetch dt.davis.problems
Davis-detected problems
timeseries avg(metric.key)
Metrics
NOT fetch — hyphenated keys need backticks: timeseries sum(my.metric-name)
smartscapeNodes "HOST"
Topology
NOT fetch — types: HOST, SERVICE, K8S_CLUSTER, etc.
dt.entity.* is deprecated — use dt.smartscape.* and smartscapeNodes for new queries.
Discover all available data objects: fetch dt.system.data_objects | fields name, display_name, type
→ references/semantic-dictionary.md for full field namespaces
samplingRatio Parameter
fetch supports a samplingRatio: parameter to reduce the volume of data read — useful for improving query performance on large datasets.
fetch spans, samplingRatio:100 // reads ~1% of data
Allowed values: depend on the concrete data object and range from 1, 10, 100, 1000, 10000 to 100000, the highest level only available for logs and spans.
Sampling is hierarchical for spans, user.events and user.sessions: a record included at a higher ratio (e.g. 100) is guaranteed to also appear at lower ratios (e.g. 10, 1), but not vice versa. This means results at different ratios are subsets of each other. All other non-metric data objects are sampled independently per record, so results at different ratios are not subsets.
The actual ratio applied is accessible via the dt.system.sampling_ratio field. Use it to extrapolate sampled counts back to true totals:
fetch logs, samplingRatio:10
| summarize count_extrapolated = sum(dt.system.sampling_ratio)
Metric Discovery
To search for available metrics by keyword, use the command metrics:
metrics from: now() - 1h
| filter contains(metric.key, "replay")
| summarize count(), by: {metric.key}
| sort `count()` desc
There is **no fetch dt.metric** or fetch dt.metrics or fetch dt.system.metrics — those data objects do not exist.
Timeseries Aggregation Functions
The timeseries command supports only these aggregation functions:
Function
Description
sum
Sum of metric data points per time slot
avg
Average of metric data points per time slot
min
Minimum of metric data points per time slot
max
Maximum of metric data points per time slot
count
Count of metric data points per time slot
percentile(metric, N)
Nth percentile per time slot. **Requires rollup:** — see below.
median(metric)
50th percentile per time slot (= percentile(metric, 50)). **Requires rollup:**.
percentRank(metric, value)
Percentile rank of a value per time slot. **Requires rollup:**.
countDistinct(metric)
Approximate distinct count per time slot (cardinality metrics only; does NOT accept rollup:).
Helpers (use alongside an aggregation): start(), end().
**Not supported by timeseries:** countIf, collectArray, stddev, variance, takeAny, takeFirst, takeLast — use summarize or makeTimeseries.
The rollup: parameter
Metrics are pre-aggregated at ingest time. rollup: controls how raw data points are combined per time slot. Required for percentile, median, percentRank — without it the query silently returns no results. avg/min/max/sum/count work without rollup:.
Single aggregation — rollup: at command level. Multiple aggregations in {} — rollup: must go inside each function call (command-level rollup: causes UNKNOWN_PARAMETER_DEFINED):
timeseries p90 = percentile(dt.process.handles.file_descriptors_percent_used, 90), rollup: avg
timeseries {
p90 = percentile(dt.process.handles.file_descriptors_percent_used, 90, rollup: avg),
med = median(dt.process.handles.file_descriptors_percent_used, rollup: avg),
avg_val = avg(dt.process.handles.file_descriptors_percent_used)
}, by: {dt.smartscape.host}
Values: avg (gauges), min, max, sum (counters), total.
Timeseries-to-scalar conversion
There are two ways to collapse a timeseries to a scalar. Prefer the scalar:true parameter when you only need the single aggregated value — it is more efficient because no array is materialized. Fall back to array functions when you need both the full series and a derived scalar in the same query.
**Preferred: scalar:true on the aggregation function**
Pass scalar:true to any timeseries aggregation function. The result field contains a single value instead of an array, and no intermediate array is allocated:
timeseries avg_cpu = avg(dt.host.cpu.usage, scalar:true), by:{dt.smartscape.host}
timeseries {
avg_cpu = avg(dt.host.cpu.usage, scalar:true),
max_cpu = max(dt.host.cpu.usage, scalar:true)
}, by:{dt.smartscape.host}
**Fallback: array functions in fieldsAdd**
When you need the full time series array alongside a derived scalar, use array functions in a subsequent | fieldsAdd:
Function
Description
arrayAvg(arr)
Average of all values in the array
arraySum(arr)
Sum of all values
arrayMin(arr)
Minimum value
arrayMax(arr)
Maximum value
arrayMedian(arr)
Median value
arrayPercentile(arr, N)
Nth percentile (0–100)
arrayLast(arr)
Last non-null value (latest data point)
arrayFirst(arr)
First non-null value (earliest data point)
timeseries cpu = avg(dt.host.cpu.usage), by:{dt.smartscape.host}
| fieldsAdd avg_cpu = arrayAvg(cpu), max_cpu = arrayMax(cpu)
Time Alignment (@-operator)
The @ operator aligns timestamps to a boundary — agents often get this wrong.
Expression
Meaning
now()@h
Current time, aligned to the hour boundary
now()@d
Midnight today
now()@w1
Monday this week
now()-2h@h
2 hours ago, aligned to the hour (offset first, then align)
Rules:
- Order: offset before alignment —
now()-2h@h, notnow()@h-2h
- No space between
@and the unit —now()@hnotnow() @h
m= minutes,M= months — do not confuse them
→ references/dql/dql-functions-timeseries.md for the full list of timeseries aggregations and rollup: rules
→ references/dql/dql-functions-array.md for arrayAvg / arrayMax / arrayPercentile / … spec
Entity & Smartscape Patterns
Entity fields are scoped per type — entity.id does not exist. Use smartscapeNodes for topology queries.
Entity
ID field in data
smartscapeNodes type
Host
dt.smartscape.host
"HOST"
Service
dt.smartscape.service
"SERVICE"
Process
dt.smartscape.process
"PROCESS"
K8s cluster
dt.smartscape.k8s_cluster
"K8S_CLUSTER"
Use toSmartscapeId() for ID conversion from strings (required!).
→ references/smartscape-topology-navigation.md
makeTimeseries Command
makeTimeseries builds a time-bucketed series from event data (logs, spans, bizevents). Unlike timeseries (which queries pre-ingested metrics), makeTimeseries aggregates data in a pipeline.
**Do not pipe timeseries directly into makeTimeseries** — it fails with INVALID_IMPLICIT_TIME_DEFAULT. To re-aggregate metric data, use start() + expand (see references/summarization.md).
fetch logs
| makeTimeseries
total = count(),
errors = countIf(loglevel == "ERROR"),
interval: 5m,
by: {k8s.cluster.name}
| fieldsAdd error_rate = errors / total * 100
Key parameters: interval:, by:{}, from:/to:, bins:, time: (timestamp field), spread: (for count/countIf only), nonempty:.
→ references/summarization.md for full makeTimeseries patterns and summarize bucketing
→ references/iterative-expressions.md for timeseries array manipulation
matchesValue() Usage
Use matchesValue() for array fields such as dt.tags:
| filter matchesValue(dt.tags, "env:production")
- Not for string fields with special characters — use
contains()for those
matchesValue()on a scalar string field does not behave like a wildcard or fuzzy match
Chained Lookup Pattern
Each lookup command without a fields parameter **removes all existing fields starting with the prefix (default: lookup.)** before adding new ones. When chaining multiple lookups, use fields parameter or custom prefixes to preserve the result:
Option 1 (default): the desired fields are known.
fetch bizevents
// Step 1: First lookup — enrich orders with product info
| lookup [fetch bizevents
| filter event.type == "product_catalog"
| fields product_id, category],
sourceField: product_id, lookupField: product_id, fields: {product_id, product_category = category}
// Step 2: Second lookup — specify fields with a different name
| lookup [fetch bizevents
| filter event.type == "warehouse_stock"
| fields category, warehouse_region],
sourceField: product_category, lookupField: category, fields: {warehouse_region, warehouse_category = category}
All 4 lookup fields product_id, product_category, warehouse_region, and warehouse_category are available.
Without the fields:{...} parameter, the fields would be prefixed with lookup. and the second lookup command would delete the fields added by the first lookup.
Option 2: keep all fields from the lookup.
fetch bizevents
// Step 1: First lookup — enrich orders with product info
| lookup [fetch bizevents
| filter event.type == "product_catalog"
| fields product_id, category],
sourceField: product_id, lookupField: product_id, prefix: "product."
// Step 2: Second lookup — specify fields with a different prefix
| lookup [fetch bizevents
| filter event.type == "warehouse_stock"
| fields category, warehouse_region],
sourceField: product_category, lookupField: category, prefix: "warehouse."
The new fields are: product.product_id, product.category, warehouse.category, warehouse.warehouse_region.
All fields starting with product. or warehouse. are removed from the original source.
Without the dedicated prefix, both lookup commands would use the same prefix (lookup.) and the second lookup drops the first lookup's results — producing empty fields.
makeTimeseries Command
makeTimeseries builds a time-bucketed series from event data (logs, spans, bizevents). Unlike timeseries (which queries pre-ingested metrics), makeTimeseries aggregates data in a pipeline.
**Do not pipe timeseries directly into makeTimeseries** — it fails with INVALID_IMPLICIT_TIME_DEFAULT. To re-aggregate metric data, use start() + expand (see references/summarization.md).
fetch logs
| makeTimeseries
{total = count(),
errors = countIf(loglevel == "ERROR")},
interval: 5m,
by: {k8s.cluster.name}
| fieldsAdd error_rate = errors[] * 100.0 / total[]
Key parameters: interval:, by:{}, from:/to:, bins:, time: (timestamp field), spread: (for count/countIf only), nonempty:. → references/dql/dql-commands.md for full spec.
Entity existence timeline using spread::
smartscapeNodes "HOST"
| makeTimeseries concurrently_existing_hosts = count(), spread: lifetime
→ references/iterative-expressions.md for timeseries array manipulation
Timeframe Specification
Access to data requires specification of a timeframe.
It can be specified in the UI, as REST API parameters, or in a DQL query explicitly using a pair of parameters: from: and to: (if one is omitted it defaults to now()), or alternatively using a single timeframe: parameter.
Timeframe can be expressed using absolute values or relative expressions vs. current time. The time alignment operator (@) can be used to round timestamps to time unit boundaries — see references/operators.md for full details.
Examples
from:now()-1h@h, to:now()@h // last complete hour
from:now()-1d@d, to:now()@d // yesterday complete
from:now()@M // this month so far, till now
from:now()-2h@h // go back 2 hours, then align to hour boundary
See references/operators.md for the full @ alignment-unit table (including m vs. M, week-day variants w1–w7, and factor rules like @3h).
Absolute timestamps
Use ISO 8601 format:
from:"2024-01-15T08:00:00Z", to:"2024-01-15T09:00:00Z"
Modifying Time
Key concepts
- DQL has 3 specialized types related to time:
- timestamp — internally kept as number of nanoseconds since epoch, but exposed as date/time in a particular timezone
- timeframe — a pair of 2 timestamps (start and end)
- duration — internally kept as number of nanoseconds, but exposed as duration scaled to a reasonable factor (e.g. ms, minutes, days)
Rules
- Subtracting timestamps yields a duration:
timestamp - timestamp → duration
- Duration divided by duration yields a double: e.g.
2h / 1m=120.0
- Scalar times duration yields a duration: e.g.
no_of_h * 1h → duration
- For extraction of time elements (hours, days of month, etc):
- ✅ Use time functions. They support calendar and time zones properly including DST.
- ❌ Avoid using
formatTimestampfor extracting time components.
- ❌ Avoid converting timestamps and durations to double/long and using division, modulo, and constants expressing time units as nanoseconds.
References
- references/useful-expressions.md — Useful expressions in DQL
- references/semantic-dictionary.md — Dynatrace Semantic Dictionary: field namespaces, data models, stability levels, query patterns, and best practices
- references/summarization.md — Various applications of summarize and makeTimeseries commands
- references/iterative-expressions.md — Array and timeseries manipulation (creation, modifications, use in filters) using DQL
- references/smartscape-topology-navigation.md — Smartscape topology navigation syntax and patterns
- references/optimization.md — DQL query optimization: filter placement, time ranges, field selection, and performance best practices
- references/operators.md —
inoperator (subquery syntax) and full@time alignment unit reference