26.4.1

Breaking Changes:

• Docker images are no longer published to Docker Hub, use the Github Container Registry instead (see documentation). (#5845)

Features:

• Use ModelMetadata config with context size and utilization. (#5814)
• Handle minidump placeholders. (#5849)
• Add config option to spread envelopes evenly across buffer partitions. (#5853)

Internal:

• Move unreal crash report expansion from processing into endpoint. (#5825)
• Ports legacy standalone span processing to the processing framework. (#5852)
• Retry failing objectstore requests. (#5836)
• Add mobile normalizations to SpanV2 processing pipeline (mobile tag, main thread, outlier filtering, app start backfill from V1 transactions, device class). (#5824)
• Remove the deprecated aiModelCosts global config, superseded by aiModelMetadata. (#5862)

Bug Fixes:

• Respond with 429 if otlp logs are rate limited, as per spec. (#5841)
• Respect rate limit silencing also when all items are rate limited. (#5840)

26.4.0

Breaking Changes:

• To prevent false positives, non-public email addresses (e.g. user@localhost) are no longer scrubbed by default. (#5737)

Bug Fixes:

• Scrub API key headers with hyphens (e.g. x-api-key) in default PII scrubbing. (#5829)
• Store segment name in sentry.transaction in addition to sentry.segment.name on OTLP spans. (#5765)
• Explicitly handle in-flight requests during shutdown. (#5746, #5769)
• Emit outcomes in both log_byte and log_item categories when logs are dropped. (#5766)
• Propagate an event's retention policy to its attachments (#5774)
• Prevent panic in span description normalization. (#5781)
• Limit overall stream size in playstation multiparts. (#5795)

Features:

• Transition error processing pipeline to a more modern implementation. (#5702)
• Set sentry.segment.id and sentry.segment.name attributes on OTLP segment spans. (#5748)
• Envelope buffer: Add option to disable flush-to-disk on shutdown. (#5751)
• Allow configuring Objectstore client auth parameters. (#5720)
• Metric size limit per metric default changed to 1mib. (#5779)
• Use gen_ai.function_id as a fallback for gen_ai.agent.name. (#5776)
• Add gen_ai.input.messages and gen_ai.output.messages as distinct fields for SpanData. (#5797)
• Merge gen_ai.request.messages into gen_ai.input.messages and gen_ai.response.text into gen_ai.output.messages. (#5813)
• Extract http.query and url.query attributes from query_string in transactions' request context. (#5784)
• Add ModelMetadata global config with context size. (#5831)

Internal:

• Update trace metric PII scrubbing to use relay_pii::eap::scrub. (#5815)
• Calculate and track accepted bytes per individual trace metric item via TraceMetricByte data category. (#5744, #5767)
• Graduate standalone span ingestion feature flag. (#5786)
• Remove cardinality limiter. (#5809)
• Use new processor architecture to process standalone profiles. (#5741)
• TUS: Disallow creation with upload. (#5734)
• Add logic to verify attachment placeholders and correctly store them. (#5747)
• Remove continuous-profiling-beta feature flags. (#5762)
• Playstation: Do not upload attachments if quota is 0. (#5770)
• Add payload byte size to trace metrics. (#5764)
• Remove transaction metrics extraction. (#5792)
• Mix kafka partition key with org id. (#5772)
• Set a trace_id on all events by default for internal use. (#5759)

26.3.1

Internal:

• Use new processor architecture to process standalone user reports. (#5731)

26.3.0

Breaking Changes:

• Use a more mature (dog)statsd backend, with proper support for unix sockets and reservoir sampling.
  Due to the new backend, all sampling and tag filtering configuration options have been removed. (#5675)
• Remove unused outcomes.emit_client_outcomes configuration option. (#5722)

Bug Fixes:

• Normalize custom metric units to none on trace metrics during processing. (#5718)
• Prevent minidump compression bomb. (#5613)
• Relay Temporarily rejected unknown headers on envelope items. (#5709)

Features:

• Populate gen_ai.response.model from gen_ai.request.model if not already set. (#5654)
• Add support for Unix domain sockets for statsd metrics. (#5668)
• Support deployment.environment OTLP resource attribute for setting the Sentry environment. (#5691)
• Allow users to opt-out of DNS caching. (#5700)
• Update device classification for new iPad and iPhone models. (#5704)
• Add TraceMetricByte data category. (#5719)

Internal:

• Strip performance metric specs from extraction while keeping extraction interfaces intact. (#5674)
• Allow deferred lengths to the /upload endpoint when the sender is trusted. (#5658)
• Stream non-prosperodmp attachments received at /playstation to objectstore. (#5673)
• Use new processor architecture to process client reports. (#5686)
• Use new processor architecture to process standalone attachments. (#5703)
• Prevent timeouts on the /upload endpoint. (#5692)
• Handle traffic bursts in the objectstore service. (#5689)
• Disable fetch_materials on GoCD pipeline-complete stages. (#5697)

26.2.1

Bug Fixes:

• Return status code 200 instead of 202 for OTLP endpoints. (#5645)

Features:

• Add experimental /upload endpoint for large file uploads. (#5638)

26.2.0

Bug Fixes:

• Emit outcomes for spans trimmed from a transaction. (#5410)
• Support sample alias in CSP reports. (#5554)
• Fix inconsistencies with Insights' expected attributes. (#5561)
• Emit outcomes for dropped standalone replay_event items. (#5634)
• Validate that EAP integer attributes fit into i64. (#5621)

Features:

• Add new frame fields for MetricKit flamegraphs. (#5539)
• Apply clock drift correction to logs and trace metrics. (#5609)
• Add Culture context to event schema. (#5615)
• Trim spans with a new EAP trimming processor. (#5616)
• Forwarded requests are now streamed instead of buffered in-memory. (#5624)

Internal:

• Add EAP double-write for session data. (#5588)
• Always process OTLP spans with the span streaming pipeline. (#5631)
• Embed AI operation type mappings into Relay. (#5555)
• Use new processor architecture to process transactions. (#5379)
• Add gen_ai_response_time_to_first_token as a SpanData attribute. (#5575)
• Add sampling to expensive envelope buffer statsd metrics. (#5576)
• Use new processor architecture to process replays. (#5580)
• Add gen_ai.cost_calculation.result metric to track AI cost calculation outcomes by integration and platform. (#5560)
• Normalizes and validates trace metric names. (#5589)
• Add manual category to cost calculation metric origin tag (#5603)
• Remove the ReplayEvents Kafka topic and the replay.relay-snuba-publishing-disabled.sample-rate option. (#5629)

26.1.0

Breaking Changes:

• Return status code 413 if a request is rejected due to size limits. (#5474)

Features:

• Updates rdkafka to 2.10 which fixes some protocol incompatibilities with Kafka 4. (#5523)
• Add sentry.category normalization for V2 spans. (#5533)
• Include cache write token cost in cost calculation for gen_ai spans. (#5530)

Bug Fixes:

• Write item IDs of logs, metrics and trace attachments in correct byte order. (#5526)
• Reworked AI span extraction to also take trace context into account. (#5515)
• Mark root spans (spans without a parent) as segments in OTEL conversion. (#5532)

Internal:

• Release Docker image to GHCR and DockerHub via Craft. (#5509)
• Tag span usage and count_per_root_project metrics with segment information. (#5511)
• Experimental support for loading configuration values from files. (#5531)
• Remove support for global quotas. (#5534)

25.12.1

Features:

• Double write to legacy attributes for backwards compatibility. (#5490)

25.12.0

Features:

• Updates Legacy Browser inbound filters. (#5326)
• Increases log size limits to better support log drains. (#5441)
• Add functionality to process and store span attachments. (#5423, #5363)
• Apply existing cookie rules to http.request.header.cookie.<key> fields. (#5456)
• Add functionality to process and store trace attachments. (#5457)
• Lower default memory utilization threshold for disk spooling from 90% to 80%. When Relay hits 80% memory utilization, it will start spooling envelopes to disk instead of processing them. (#5472)

Bug Fixes:

• Fix parsing of data categories/quotas when using an aliased data category name. (#5435)

Internal:

• Revise trace metric and log size limits. (#5440)
• Update is_ai_span and infer_ai_operation_type to use gen_ai.operation.name. (#5433)
• Add project_id to profile item kafka headers. (#5458)
• Remove recycle_check_frequency from Redis configuration. (#5476)
• Remove gen_ai_usage_total_cost attribute and stop double writing costs. (#5471)

25.11.1

Breaking Changes:

• Switch Statsd histogram metrics to distribution metrics. (#5378)

Features:

• Support comparing release versions without build code. (#5376)
• Support uploading attachments directly to objectstore. (#5367)
• Add span_count item header to the envelope protocol. (#5392)
• Add event.name attribute to OTLP logs. (#5396)
• Remove feature flag for Vercel Log Drain endpoint. (#5406)
• Add http.request.header.cookie to fields scrubbed by cookie rules. (#5408)
• Map request_id to trace_id in vercel log drain transform. (#5333)
• Populate sentry.platform during OTLP ingestion. (#5411)
• Implement optional opportunistic caching for quotas. (#5416)
• Add db attribute normalization to V2 spans. (#5399)

Internal:

• Derive the rate limiting decision in Relay from consumed quota. (#5390)