The OpenTelemetry Collector Contrib contains everything in the opentelemetry-collector release, be sure to check the release notes there as well.

End User Changelog

🛑 Breaking changes 🛑

• cmd/opampsupervisor: Remove the reports_package_statuses capability config option. The accepts_packages option now enables both the AcceptsPackages and ReportsPackageStatuses OpAMP capabilities. (#49762) Neither capability was functional; the supervisor exits with an error at startup when configured, so no working configuration is affected.

• processor/drain: Add masking rules for named parameter extraction. Removes the extract_parameters and params_attribute config fields released in v0.157.0. (#48914) Adds masking_rules, an ordered list of {name, pattern} regex substitutions applied to a copy of the log body before it is fed to the Drain tree. Matched substrings become named mask tokens in derived templates (for example <ip>), stabilising the tree on high-cardinality values. Each masked position writes a dynamic attribute at <parameter_key_prefix>.<mask name> (default prefix log.record.template.parameter), matching the OTel semantic-convention pattern used by http.request.header.<key> and db.query.parameter.<key>.

  Adds emit_wildcards (default false): when true, Drain's own <*> positions are written to a positional string slice attribute at wildcards_attribute (default log.record.template.wildcards). Independent of masking_rules, so users can enable it without any rules to see raw variable values before deciding what to mask.

  When a mask name matches multiple positions in the same template, first-match wins and the losing values are dropped. The otelcol_processor_drain_masks_duplicates counter is incremented once per record per duplicated mask name, tagged with a mask attribute for observability.

  Breaking: the positional-only extract_parameters and params_attribute fields released in v0.157.0 are removed. Collectors with either field set will fail to start after upgrading. To migrate, replace extract_parameters: true with emit_wildcards: true, and rename params_attribute to wildcards_attribute. The behaviour is identical; only the field names change. Note the default output attribute also renames from log.record.template.params to log.record.template.wildcards, so downstream consumers of the old attribute key should either update their references or set wildcards_attribute: log.record.template.params to keep the old key.

• processor/dynamic_sampling: Flatten sampler config, rename key_fields to key_attributes, drop initial_sampling_rate. (#49311) Sampler fields no longer live under a per-type sub-block (sampler.<type>.<field>); they move up one level under sampler: with type acting as the discriminator. key_fields is renamed to key_attributes to match OTel attribute vocabulary used elsewhere in the processor. The rarely-used initial_sampling_rate on ema_throughput is removed; dynsampler-go's default applies. goal_throughput_per_sec is now int for both throughput samplers. Fields set for a sampler type that does not use them are rejected at config validation.

• processor/dynamic_sampling: Replace the custom rule condition parser with OTTL expressions and add a match field to control same-span vs any-span semantics. (#49311) Rule conditions: are now OTTL boolean expressions evaluated in the ottlspan context. Path expressions must be qualified with a context prefix (span.attributes["k"], resource.attributes["k"], span.status.code, etc). A new per-rule match: field selects any_span (default, each condition satisfied by some span) or same_span (some single span satisfies all conditions). A new counter, otelcol_processor_dynamic_sampling_ottl_eval_errors, is labelled by rule and tracks runtime OTTL evaluation errors.

• receiver/icmpcheckreceiver: Change RTT metric value type from int to double for sub-millisecond precision (#49960)

• receiver/oracledb: oracle.db.pdb is now a data-point attribute (opt-in) instead of a resource attribute. (#48643) oracle.db.pdb has moved from a resource attribute to an opt-in data point attribute. Downstream pipelines that grouped, routed, or filtered on oracle.db.pdb at resource scope must be updated to read it from data point attributes, and it must be explicitly enabled on each metric that should carry it via metrics.<name>.attributes: [oracle.db.pdb]. Existing CDB deployments that do not enable the attribute or add the grants needed for per-PDB collection keep working unchanged; the receiver falls back to the single-container queries.

• receiver/postgresql: Collect postgresql.database.locks from each configured database instead of only the default postgres database. (#49206) Previously the metric was collected once against the default postgres database, so locks on relations belonging to other configured databases were silently dropped, and all data points were emitted on the instance-level resource without postgresql.database.name. Lock data points for database-local relations are now emitted on the corresponding per-database resource with the postgresql.database.name resource attribute. Locks on shared system catalogs (e.g. pg_database) are reported once at the instance level. When the receiver.postgresql.useOTelSemconv feature gate is enabled there is a single server-level resource, so the data points instead carry a db.namespace attribute to identify the database. Without it relations that exist in more than one database (any system catalog, or user tables sharing a name) would collapse into a single series. Locks on shared system catalogs carry an empty db.namespace, since they are server-scoped rather than database-scoped. The lock count now uses COUNT(*) instead of COUNT(pid), so locks held by prepared transactions (which have a NULL pid in pg_locks) are counted instead of being reported as zero. The metric is disabled by default and has development stability, so no feature gate is provided for this behavior change.

🚩 Deprecations 🚩

• exporter/mezmo: Deprecate the mezmo exporter (#49953) Mezmo now supports ingesting OpenTelemetry data directly via OTLP. Use the OTLP/HTTP exporter instead. See https://docs.mezmo.com/telemetry-pipelines/otel-collector and https://docs.mezmo.com/telemetry-pipelines/open-telemetry-source for migration guidance.

• processor/alibabaecsdetector: Deprecate per-detector fail_on_missing_metadata in the alibaba_ecs detector config (#46579) Use the top-level fail_on_missing_metadata in the processor config instead. The field continues to work but will emit a deprecation warning in the logs when set and code will be removed later.

• processor/delta_to_cumulative: Rename the 'deltatocumulative' processor to 'delta_to_cumulative'. The old 'deltatocumulative' type remains available as a deprecated alias. (#45339)

• processor/delta_to_rate: Rename the 'deltatorate' processor to 'delta_to_rate'. The old 'deltatorate' type remains available as a deprecated alias. (#45339)

• processor/novadetector: Deprecate per-detector fail_on_missing_metadata in the nova (OpenStack) detector config (#46579) Use the top-level fail_on_missing_metadata in the processor config instead. The field continues to work but will emit a deprecation warning in the logs when set and code will be removed later.

• processor/resource_detection: Deprecate per-detector fail_on_missing_metadata in the ec2 detector config (#46579) Use the top-level fail_on_missing_metadata in the processor config instead. The field continues to work but will emit a deprecation warning in the logs when set and code will be removed later.

• processor/tencentcvmdetector: Deprecate per-detector fail_on_missing_metadata in the tencent_cvm detector config (#46579) Use the top-level fail_on_missing_metadata in the processor config instead. The field continues to work but will emit a deprecation warning in the logs when set and code will be removed later.

• processor/upclouddetector: Deprecate per-detector fail_on_missing_metadata in the upcloud detector config (#46579) Use the top-level fail_on_missing_metadata in the processor config instead. The field continues to work but will emit a deprecation warning in the logs when set and code will be removed later.

• processor/vultrdetector: Deprecate per-detector fail_on_missing_metadata in the vultr detector config (#46579) Use the top-level fail_on_missing_metadata in the processor config instead. The field continues to work but will emit a deprecation warning in the logs when set and code will be removed later.

🚀 New components 🚀

• extension/aws_iam_db_auth: Add an extension awsiamdbauthextension that implements dbauth. (#49044)
• receiver/dns_check: Add initial skeleton of DNS Check receiver (README, config, factory, metadata) with In Development stability. (#49561)

💡 Enhancements 💡

• exporter/datadog: Add exporter.datadogexporter.AddUnits feature gate that maps OTLP (UCUM) metric units to their Datadog equivalents. (#15280)

• exporter/elasticsearch: Allow traces, profiles, and synthetics as valid data_stream.type values when overriding via attributes in bodymap mapping mode, in addition to the existing logs and metrics. (#49337)

• exporter/prometheus_remote_write: Add convert_explicit_histograms_to_nhcb to convert explicit-bucket (classic) histograms into Native Histograms with Custom Buckets (NHCB) on export, with keep_classic_histograms to emit both representations during migration. (#33661) When convert_explicit_histograms_to_nhcb is set, each OTLP explicit-bucket histogram is converted to a single NHCB series (schema -53) carrying the bounds as CustomValues, instead of the classic _bucket/_sum/_count fan-out. Setting keep_classic_histograms: true additionally emits the classic series so a migration can run both in parallel. Implemented for both the RW1 and RW2 write paths; reuses Prometheus' util/convertnhcb converter for wire-compatible encoding.

• exporter/prometheus_remote_write: Add include_metadata_keys to remote_write_queue configuration to forward client metadata as HTTP headers. (#47317)

• pkg/jaeger: Migrate http.status_code (v1.25.0) semantic convention to http.response.status_code (v1.40.0) (#45036) The migration is gated behind two alpha (disabled by default) feature gates: enable pkg.translator.jaeger.EmitV1HttpConventions to emit http.response.status_code, and additionally enable pkg.translator.jaeger.DontEmitV0HttpConventions to stop emitting the deprecated http.status_code. Both gates will graduate to beta (enabled by default) and eventually be removed, after which the translator will only emit http.response.status_code. That removal is a breaking change for anything still relying on http.status_code, so please migrate your dashboards, alerts and queries now.

• pkg/stanza: Change connection_idle_timeout on the tcp_input operator to a duration field that defaults to no idle timeout, apply it independently of max_connections, and add a tcp_input_refused_connections metric. (#49610)

• pkg/stanza: Add max_connections and connection_idle_timeout options to the tcp_input operator to limit the number of concurrent TCP connections. (#49610)

• processor/dynamic_sampling: Evicted traces now receive a real sampling decision instead of being dropped silently, with a configurable eviction policy. (#49311) When num_traces is full the oldest pending trace is evicted and decided immediately. eviction.policy: evaluate (default) runs the normal rules on the spans seen so far; eviction.policy: probabilistic decides with a threshold derived from eviction.sampling_percentage in constant time, shedding load under pressure. Both modes record the decision for late spans and stamp kept traces with a correct ot=th.

• processor/dynamic_sampling: Add configurable root_span_condition OTTL expression to control which spans trigger the trace decision. (#49311) Defaults to IsRootSpan(), preserving prior behaviour. Operators can override with any OTTL boolean expression evaluated in the ottlspan context (e.g. accepting a producer-supplied hint attribute or a cross-process server span). Evaluation errors are counted on processor_dynamic_sampling_ottl_eval_errors under the sentinel rule="_root_span_condition" label.

• processor/oracleclouddetector: Add cloud.resource_id resource attribute to the Oracle Cloud detector, set to the compute instance OCID. (#49832)

• processor/resource_detection: Add top-level fail_on_missing_metadata to make unreachable metadata services a hard failure (#46579) When true, network-based detectors return an error instead of silently returning an empty resource when their metadata service is unreachable. Supersedes the per-detector fail_on_missing_metadata fields, which are now deprecated.

• processor/resource_detection: Add cloud.region to GKE resource detection by deriving it from cloud.availability_zone (#49694)

• processor/span_pruning: Preserve whole outlier subtrees instead of individual spans and detect outliers at every aggregation level, so a slow interior span (e.g. one slow handler among many) keeps its entire subtree. (#49324)

• processor/span_pruning: Add optional OTTL conditions filtering so span pruning only applies to traces with at least one matching span (empty conditions still prune all traces), and add the otelcol_processor_spanpruning_traces_skipped metric for traces skipped when no conditions match. (#49026)

• processor/transform: Add ParseELF function to parse W3C Extended Log Format (ELF) log blocks into structured maps. (#48352) ParseELF(target) parses a complete ELF text block and returns a pcommon.Map with directive metadata (version, software, date, start_date, end_date, remark), a fields slice, and an entries slice keyed by field name. Multiple #Fields directives and double-quoted values (IIS-style) are supported.

• receiver/awsxray: Migrates v1.20.0 (http.client_ip) semantic convention to v1.42.0 (client.address) (#45085)

• receiver/awsxray: Migrated HTTP semantic conventions from v1.25.0 to v1.42.0 (#45084)

• receiver/kafka_metrics: Add kafka.cluster.id resource attribute, auto-discovered from cluster metadata. Disabled by default; opt in via resource_attributes. (#48892) The attribute is disabled by default. When enabled, it complements the existing user-configured kafka.cluster.alias resource attribute.

• receiver/kubelet_stats: Add the receiver.kubeletstats.cpuUsageScrapeBased feature gate. When enabled, container.cpu.usage, k8s.pod.cpu.usage and k8s.node.cpu.usage (and the cpu utilization metrics derived from them) are calculated as the rate of the corresponding *.cpu.time counter between consecutive scrapes, instead of being read directly from the kubelet's UsageNanoCores value. (#49477)

• receiver/memcached: Add tls configuration to support connecting to memcached over TLS. (#49146) TLS is disabled by default (insecure: true), so existing plaintext configurations are unaffected.

• receiver/mongodb: Add db.server.top_query slow query event collection to the MongoDB receiver. (#49623) Emits the top N slowest query executions per scrape window as db.server.top_query log events, including obfuscated query text, execution stats, and optional explain plans. Configurable via the new top_query_collection config block. The logs signal is at development stability; attribute names may change until OTel db.server.top_query conventions stabilize.

• receiver/mongodb: Add service.name and service.namespace opt-in resource attributes and allow overriding any resource attribute via override_value. (#49812) When service.name is enabled, the receiver sets it to unknown_service:mongodb per OTel specification. When service.namespace is enabled, it defaults to an empty string until set via configuration. Each resource attribute now accepts an override_value under resource_attributes, letting users pin values such as service.name, service.namespace, or service.instance.id to uniquely identify database instances across environments.

• receiver/netflow: Add interface, IP header, L2, ICMP, routing, and IPFIX observation attributes to parsed flow log records (#49946) Added the following attributes decoded by goflow2 but previously not included in log records: flow.in_if, flow.out_if, flow.ip_tos, flow.ip_ttl, flow.ip_flags, flow.fragment_id, flow.fragment_offset, flow.ipv6_flow_label, flow.icmp_type, flow.icmp_code, flow.src_mac, flow.dst_mac, flow.src_vlan, flow.dst_vlan, flow.vlan_id, flow.next_hop, flow.next_hop_as, flow.src_as, flow.dst_as, flow.bgp_next_hop, flow.src_net, flow.dst_net, flow.forwarding_status, flow.observation_domain_id, flow.observation_point_id.

• receiver/oracledb: Add real-time workload rate metrics. (#49749)

• receiver/oracledb: Speed up query-sample collection by splitting it into two queries so the SQL text/plan lookup no longer joins V$SESSION against the entire cursor cache. (#49874) The session query no longer joins V$SQL (which forced Oracle to scan the whole shared-pool cursor cache and materialize SQL_FULLTEXT for every cursor). Instead, the receiver collects the active sessions first, then fetches SQL_FULLTEXT / CHILD_ADDRESS / PLAN_HASH_VALUE from V$SQL for only those sql_ids via WHERE SQL_ID IN (...), and joins the results in the collector. Sessions whose cursor has aged out of the shared pool are skipped, preserving the previous inner-join semantics. Note: the two queries are issued at slightly different SCNs; for actively-executing cursors this is immaterial (they are pinned in the shared pool), but cursors that age out between Pass-1 and Pass-2 will not appear in the output, which matches the previous inner-join semantics.

• receiver/oracledb: Add PDB auto-discovery and per-PDB metrics for Oracle multitenant (CDB) deployments via a single CDB root connection. (#48643) When connected to an Oracle CDB root (Oracle 12c+), the receiver automatically detects all PDBs and tags per-PDB metrics with the opt-in oracle.db.pdb data point attribute. The attribute is also populated for direct-PDB connections so the metric hierarchy is consistent regardless of how the collector connects. Non-CDB instances and Oracle <12c are unaffected (the attribute is left empty).

• receiver/oracledb: Add real-time I/O rate metrics. (#49748) All metrics are disabled by default with development stability.

• receiver/postgresql: Add the optional postgresql.query.execution.time metric. (#49822)

• receiver/postgresql: Adopt dbauth config in postgresql receiver. (#49044)

• receiver/postgresql: Add opt-in pgvector metrics. (#49576) Adds opt-in metrics for pgvector similarity-search and insert activity, all disabled by default.

• receiver/receiver_creator: Add support for os detection in rules (#49975)

   receiver_creator:
   watch_observers: [host_observer]
   receivers:
     windows_service:
       # Enable this receiver if the OS is Windows.
       rule: type == "hostport" && os == "windows"
       config:
         include_services:
           - MSSQLSERVER
         collection_interval: 10s

• receiver/redis: Add pub/sub metrics (redis.pubsub.channels, redis.pubsub.patterns, redis.pubsub.shard_channels, redis.pubsub.clients) scraped from Redis INFO stats fields. (#49147)

• receiver/sqlserver: Add opt-in metrics for monitoring SQL Server Always On Availability Group database replicas. (#49633)

  • sqlserver.availability_group.database_replica.secondary_lag
  • sqlserver.availability_group.database_replica.queue.size
  • sqlserver.availability_group.database_replica.queue.rate

• receiver/sqlserver: Add opt-in metrics for monitoring host-level CPU, memory, and disk I/O as observed by SQL Server. (#49862)

  • sqlserver.cpu.utilization
  • sqlserver.host.memory.limit
  • sqlserver.host.memory.usage
  • sqlserver.disk.io
  • sqlserver.disk.operations

• receiver/sqlserver: Support server properties query on Azure SQL Managed Instance (#49444)

• receiver/udp_log: Add the stanza.udp.useStableNetworkAttributes feature gate to emit stable network semantic convention attributes when add_attributes is enabled. (#49050) When the stanza.udp.useStableNetworkAttributes feature gate is enabled, the add_attributes option emits the stable network attributes (network.transport, network.local.address, server.port, server.address, network.peer.address, client.port, client.address) instead of the deprecated ones (net.transport, net.host.ip, net.host.port, net.host.name, net.peer.ip, net.peer.port, net.peer.name).

🧰 Bug fixes 🧰

• exporter/elasticsearch: Fix Elasticsearch retry attempts incorrectly sharing a single timeout deadline (#45747) The configured timeout now applies independently to each HTTP request attempt.

• exporter/load_balancing: Fix a memory leak in the Kubernetes resolver where pod hostnames were retained indefinitely after pods churned when return_hostnames is enabled. (#49757) During a rolling update a pod frequently appears in an EndpointSlice a moment before its Hostname field is populated. With return_hostnames: true, the resolver previously discarded the entire update whenever any endpoint in the slice lacked a hostname, so pods that churned out in that same event were never removed from the endpoint store. Over many pod rolls the store, the hash ring, and the per-endpoint exporter map grew without bound. Endpoints missing a hostname are now skipped individually while the rest of the slice is still processed, so churned-out pods are removed promptly.

• extension/text_encoding: Fix a bug in the text encoding extension where logs were silently truncated if the input had more than 1000 records. (#49818)

• pkg/datadog: Fix agentcomponents.WithAPIConfig silently discarding its configured api_key/site in favor of DD_API_KEY/DD_SITE environment variables when they are present in the process environment. (#49957)

• pkg/stanza: Discard partial log lines instead of emitting them as truncated entries when the TCP input is shut down mid-transmission. (#49622) On graceful shutdown the operator force-closes in-flight connections. Previously any partial (non-delimited) data left in the receive buffer was flushed as a complete log entry, producing truncated records. It is now discarded, while a final line without a trailing delimiter is still emitted when the client closes the connection cleanly.

• pkg/winperfcounters: Fix batch scrape failures caused by transient PDH errors during wildcard queries. (#49416) Skips performance counter instances that return PDH_INVALID_DATA, PDH_NO_DATA, or PDH_CALC_NEGATIVE_DENOMINATOR instead of failing the entire metric batch. This prevents errors when monitoring ephemeral processes.

• processor/redaction: Apply blocked_values patterns in the order they are listed in the configuration instead of a nondeterministic order (#49858) Previously the patterns were applied in Go map iteration order. When two patterns could match overlapping parts of the same value, the result changed from run to run, and some orders left data unmasked that another order would have redacted.

• receiver/azure_monitor: Fix metric data loss and incorrect timestamps (#49532)

• receiver/datadog: Add support for the v3 series intake (/api/intake/metrics/v3/series), the default metrics endpoint for Datadog Agent 7.81.0+ (#49698) Datadog Agent 7.81.0 and 7.81.1 submit series metrics to /api/intake/metrics/v3/series by default. The receiver had no handler for that path, so series from those agents fell through to the catch-all route and were acknowledged with 200 OK but silently discarded. The v3 payload is a columnar, dictionary-encoded protobuf format distinct from v2; it is now decoded and translated to OTLP through the same path as v2 series. The v3beta route used by shadow/validation traffic is also handled.

• receiver/elasticsearch: Report elasticsearch.cluster.state_queue with state: pending using the pending count instead of the committed count (#49652) The data point for the pending state was recorded from the committed queue count, so it always mirrored the committed data point. It now uses the pending count from the node discovery stats.

• receiver/fluent_forward: Delay Fluent Forward chunk acknowledgments until logs are successfully consumed downstream. (#46973)

• receiver/googlecloudmonitoring: CUMULATIVE metrics from Cloud Monitoring will now be properly marked as monotonic upon conversion (#49804)

• receiver/postgresql: Fixes a bug in explainQuery so that it honors context cancellation (#49632) The DEALLOCATE PREPARE cleanup now runs on a detached, time-bounded context so prepared statements are still released from pooled connections even when the scrape context is canceled.

• receiver/sqlserver: Populate service.instance.id in Windows Performance Counter mode. (#49878) Previously, service.instance.id was only set by the direct connection scraper and was absent from all metrics emitted in Windows Performance Counter mode, despite being enabled: true in the default config. The Windows PC scraper now computes service.instance.id at initialization using the same logic as the direct connection scraper: <computer_name>:1433 when computer_name is configured (remote monitoring), or <os.Hostname()>:1433 when monitoring the local machine. Port 1433 is used as a default since Windows Performance Counter mode does not establish a TCP connection.

• scraper/nfs: Linux implementation no longer errors if NFS /proc files are unavailable, as this is expected when the client and/or server is inactive (#49688)

API Changelog

🛑 Breaking changes 🛑

• all: Stops embedding all configuration fields (#49797, #49913, #49914, #49915, #49965, #49966, #49972, #49973, #49974)

  • Embedded fields do not work well with tooling and libraries such as mapstructure and should be avoided.

• pkg/ottl: OTTL context path setters now handle nil values based on the path type (#49728)

We are thrilled to welcome our first-time contributors to this project. Thank you for your contributions @bgola-signalfx, @nmromero4792, @srstrickland, @rajrohanyadav, @skreuzer, @LucasHocker, @avadla, @somaz94, @MenakaDev, @krshn-ptl, @yamoyamoto, @gizas, @sarika-03, @Krishnachaitanyakc, @swapnil-signoz ! 🎉