publish: getsentry/relay@26.10.0 #9867

Breaking Changes:

• Stop producing generic metrics (spans/transactions metric buckets) to Kafka and remove the
  metrics_generic / ingest-performance-metrics topic. Session (release health) metrics are
  unchanged. (#6388)
• Stop accepting metrics in statsd format. (#6428)
• Stop parsing on-demand metrics extraction config. All extracted metrics are now hard-coded. (#6439)

Features:

• Extend cookie scrubbing to response cookies. (#6391)
• Scrub numbered variants of sensitive cookies. (#6392)
• Communicate desired chunk size to clients for tus uploads. (#6394, #6504)
• Implement support for SDK indicated requests on localhost for inbound filters. (#6422)
• Add a cidr rule condition that matches IP addresses against a list of addresses and CIDR ranges, and expose the envelope's client IP to generic inbound filters as envelope.client_ip. (#6374)
• Add a release rule condition that compares the version of a release with eq, gt, gte, lt, or lte, so generic inbound filters can express version ranges. (#6430)
• Copy the browser navigation type and id to web vital metrics. (#6401)
• Disable dynamic sampling and metrics extraction in Managed relays. For users with a dynamic sampling rate < 1, this increases the outgoing request volume of the Managed relay. (#6413, #6425)
• Copy the INP interaction type to the INP web vital metric. (#6420)
• Extract session updates from Switch dying messages. (#6363)
• Add spool.envelopes.max_unspool_envelopes_per_second to limit unspooling rate. Disabled by default. (#6433)
• Introduce dimension-based rate-limiting. (#6387)
• Add file_revision field to the stack frame schema. (#6451)
• Infer the memory AI operation type from gen_ai memory operations (create_memory, create_memory_store, delete_memory, delete_memory_store, search_memory, update_memory, upsert_memory) and the gen_ai.memory.client span op. (#6480)
• Serialize request, trace context and metadata from transaction events into segment spans. (#6490)
• Span name and description inference now support array attributes. (#6482)
• Added support for Integrity policy ingestion, not publicly accessible. (#6465, #6512)
• Add dimensions to rate limit caches. (#6479)

Bug Fixes:

• Stop inferring the bare gen_ai span operation from gen_ai.provider.name. (#6474)
• Align TUS implementation with spec. (#6408)
• Drop spans if they have an invalid sentry.segment_id. (#6414)
• Add the browser name and version to web vital metrics. (#6406)
• Limit the number of deserialization ops on ItemContainer. (#6429)
• Don't cache dimensioned rate-limits. (#6457)
• Slugify monitor slugs the same way as Sentry for check-in routing and rate limit dimensions. (#6456)
• Preserve the client exception thread ID, handled status, and fallback stack when preparing minidump events. (#6464)
• Bump symbolic so Unreal crash parsing no longer accepts arbitrarily large custom nodes in RuntimeProperties. (#6470)
• Fix accidental value duplication for large metrics buckets. (#6471)
• Delay shutdown until ephemeral disk spool has been drained. (#6484)
• Add more sensitive session cookie names for PII scrubbing. (#6499)

Internal:

• Update sentry-conventions to 0.27.0. Adds Anthropic tool-result, NEL sampling, integrity policy, AWS, MCP, and gen_ai memory conventions, along with span op deprecations. (#6498)
• Accept the projects:relay-automatic-json-expansion flag in project configs. (#6459)
• Add a flush timeout to the envelope buffer's in-memory queues. (#6375)
• Serialize patterns in their normalized forms instead of the original pattern. (#6442)
• Update the Sentry Rust SDK to 0.49 and use its sentry-minidump integration for the crash handler. (#6405)
• Use the Arroyo Kafka producer backend for processing mode. This changes the names and tags on Kafka stats reported via statsd. (#6383, #6396, #6397, #6409)
• Support resumable uploads (feature-flagged). (#6421, #6502)
• Remove unused configuration options outcomes.batch_size and outcomes.batch_interval. (#6400)
• Remove long deprecated TLS settings from Relay's configuration. (#6434)
• Update sentry-conventions to 0.25.0. server_name now backfills into server.address instead of device.name, legacy gen_ai cache token attributes backfill into their replacements, and span names and descriptions are now inferred for ui.* operations. Also adds the sentry.link.type span link attribute and a deprecated replay_id search alias for sentry.replay_id. (#6417, #6445)
  Switch default PyPI index to Socket Firewall. (#6460)