Skip to content

Changelog

1.22.2 (2026/11/10)

New Features

  • Added optional tail sampling spill. When the memory budget is insufficient, groups pending decision are spilled to disk. This supports resource limits, usage and error metrics, and Sinker routing by point labels (!234, !235).
  • Added spill configuration, capacity estimation, Kubernetes storage deployment, and troubleshooting guides (!236).

Bug Fixes

  • Fixed an issue in legacy raw tail sampling where, after recompression of small packets, data could be received successfully but not delivered downstream (!235).
  • Fixed an issue in tail sampling spill multi-target routing where continuous retries to failed targets blocked other targets in the same group; successful targets are not sent repeatedly (#138).

1.22.1 (2026/11/10)

New Features

  • Added metrics for tail sampling stage duration, send resident bytes and budget, and bytes before and after compression (#135).

Feature Optimization

  • Optimized tail sampling batch sending and buffer reuse, reducing duplicate allocations and copies during decompression, batching, and sending (#133).
  • Reduced request processing overhead when HTTP access logs are disabled, applicable to the gin_log: "" configuration (#133).

Compatibility Notes

  • Downstream requests for direct tail sampling and Sinker routing now default to zstd instead of gzip, with no automatic negotiation or fallback. Before upgrading, ensure that all downstream KodoX Point Workers support zstd; cache replay preserves the original encoding and request body (#133).

1.22.0 (2026/11/09)

Bug Fixes

  • Upgraded Go to 1.26.10 to fix a certificate trust security issue in crypto/x509 (CVE-2026-111943 / GO-2026-6809) (#130).

Feature Optimization

  • Optimized tracing tail sampling performance when processing multiple shards of the same trace (#129).

Compatibility Notes

  • The base OS of the main image was upgraded from Ubuntu 22.04 to 24.04, supporting linux/amd64 and linux/arm64 (#131).

1.21.0 (2026/10/14)

New Features

  • Added OpenLineage data ingestion, gzip forwarding, and Sinker routing (#112, #122).
  • Added the OCSF receiving endpoint, which supports batch data and writes to the ocsf Log Index (#120).

Bug Fixes

  • Fixed an issue where invalid aggregation parameters could cause process exceptions (#117).
  • Fixed an issue where stalled uploads occupied concurrency slots for extended periods and proxy response status codes were incorrect (#118).

Feature Optimization

  • Improved HTTP timeout control to prevent stalled requests from occupying resources for extended periods (#118).
  • Standardized token redaction in logs, retaining the first 8 and last 4 characters and fully hiding short tokens (#121).

Compatibility Notes

  • Downstream 4xx responses retain their original status codes. If the error body cannot be parsed, {} is returned instead of 500 (#124).

1.20.0 (2026/09/10)

New Features

  • Added an HTTP concurrency limit before reading the Body, reusing kodo_queue.workers (DW_KODO_QUEUE_WORKERS, default 256). After the limit is reached, at most 128 additional requests can wait, for up to 1s. The limit remains effective after the Kodo dispatch queue is disabled, and requests without a Body are not counted (#115).
  • Added the dataway_http_concurrency_in_flight, dataway_http_concurrency_waiting, dataway_http_concurrency_requests_total, and dataway_http_concurrency_wait_seconds metrics to observe in-flight requests, waiting requests, admission results, and wait duration (#115).

Bug Fixes

  • Fixed a compatibility issue with Langfuse SDK 4.x when /api/public/ingestion succeeds upstream but the response body is empty: return successes/errors JSON by event ID in the request batch and preserve existing JSON response content from upstream (#116).

Compatibility Notes

  • When the HTTP concurrency wait queue is full or the wait times out, registered DK-compatible endpoints return 503 and other endpoints return 429, all with Retry-After: 1; these rejected requests do not enter the disk cache (#115).
  • When an HTTP/1.x request is rejected by the concurrency limit, terminate reading the remaining Body and close the connection after returning the response, avoiding waiting for unfinished uploads; clients need to establish a new connection when retrying (#115).

1.19.0 (2026/08/27)

New Features

  • Added the /v2/tail_sampling zstd reporting endpoint and compatibility with zstd packets sent by DataKit 2.10 via /v1/tail_sampling without a compression negotiation header (#110).
  • Added bounded batch sending and failure retries for tail-sampled retained data, changed memory overflow spill-to-disk to asynchronous processing, and improved restart recovery and shutdown cleanup mechanisms (#110).

Feature Optimization

  • Tail sampling extracts trace_id after decompression and recalculates predicates; it uses gzip compression when batch sending downstream, reducing request count and transfer overhead (#110).
  • Improved recording coverage for runtime metrics such as tail sampling queue submission results, wait duration, disk depth, and batch status (#110).

Security and Observability

  • Set token labels in metrics to redacted, no longer retaining original token fragments, to avoid credential leaks and high-cardinality time series (#90).
  • Redacted errors from aggregation, tail sampling, and multi-backend forwarding to no longer expose backend URLs, response content, or authentication information (#90).

1.18.0 (2026/08/12)

Feature Optimization

  • Optimized the trace_id extraction path for tail sampling to avoid repeated payload traversal when RawGroupId is available while keeping sampling behavior unchanged (#108).
  • Upgraded cliutils to automatically initialize capacity metrics after the disk cache is opened, and improved close and file lock release behavior (#111).

Bug Fixes

  • Unified token validation and propagation semantics for special Aggregate and Tail Sampling APIs (#88).
  • Isolated the invalid token negative cache by token, write route, and main/headless mode to avoid cross-scenario false blocking (#106).
  • Fixed concurrent access issues in the failure cache and tail sampling disk degradation paths, avoiding data races during disk queue initialization, reads, and shutdown (#111).
  • Isolated pprof and Prometheus HTTP mux; pprof is disabled by default, and when enabled, it can only listen on loopback, with limits on diagnostic request duration and concurrency (#89).
  • Redacted tokens, credentials, global tags, and rule content in request forwarding, cache replay, aggregation, tail sampling, and Sinker logs (#104).
  • Fixed system Workspace URL determination and added regression verification for token injection in normal and cascading modes (#104).
  • Corrected main HTTP Server request header timeout, Gin mode, IP blocklist configuration examples, and API wrapper error handling (#104).

Security and Observability

  • Pinned the Go and golangci-lint download packages used by CI, verified SHA-256 by architecture, and removed the direct execution path for remote installation scripts (#101).
  • Updated the default DataWay dashboard, adding queue, tail sampling, Sinker cache, and runtime monitoring (#85).

1.17.0 (2026/07/14)

New Features

  • Added OTLP HTTP/protobuf endpoints: /otel/v1/traces, /otel/v1/metrics, and /otel/v1/logs (#81).
  • OTLP requests support query token, X-Token, and Bearer token, and are integrated with the Kodo queue and disk cache (#81).
  • OTLP requests support pass-through of Kodo status codes, body, and required headers (#81).

Feature Optimization

  • Optimized Sinker routing for aggregation and tail sampling so that a failure in one group no longer affects other groups (#71).
  • The invalid token negative cache applies only to eligible write requests (#77).
  • Optimized aggregation memory usage and concurrency stability.
  • Optimized tail sampling backlog handling; when the memory queue is full, data can be temporarily stored to disk.
  • Improved stability in disk cache exception scenarios.

Bug Fixes

  • Fixed race and capacity issues in the Sinker rule cache (#93).

Observability

  • Added file descriptor metrics dataway_open_files_by_type and TCP metrics dataway_tcp_connections.
  • Added tail sampling, send queue, worker, and disk staging metrics.

1.16.1 (2026/07/09)

Bug Fixes

  • /v1/write/rum rejects empty bodies and returns 400 and dataway.emptyBody; non-empty bodies are still forwarded as-is (#82).
  • When the Kodo queue is full, cacheable requests are written to disk; when the cache is unavailable, 503 is returned (#83).
  • Return "cached" only after a successful write, and improved recording of cache failure reasons (#83).

1.16.0 (2026/07/02)

New Features

  • Added IPv6 and dual-stack deployment support; the IP allowlist supports IPv4, IPv6, and CIDR (#79).
  • Added bind_ipv6/DW_BIND_IPV6 and trusted_proxies/DW_TRUSTED_PROXIES configurations (#78, #79).

Feature Optimization

  • When the Kodo success response is empty, no longer return an empty body and Content-Type (#78).
  • Return CORS headers only for requests with Origin, and add X-Content-Type-Options: nosniff to forwarded responses (#78).

1.15.0 (2026/05/27)

New Features

  • Added a bounded Kodo queue covering write, upload, and Firehose requests.
  • The queue is enabled by default and supports DW_KODO_QUEUE_* environment variables.
  • When the queue is full, cacheable requests are cached. When the cache is unavailable, 503 is returned.
  • Added an invalid token negative cache, supporting DW_TOKEN_NEGATIVE_CACHE_* environment variables.
  • The negative cache has a default TTL of 5m and stores up to 1000 tokens.

Feature Optimization

  • Added token and payload validation to the aggregation endpoint.
  • Aggregation sending now uses a background queue, supports 3 retries, and adds Content-MD5.
  • The aggregation proxy reuses the reverse proxy and validates related configuration.
  • Cache replay uses the current remote_host, and the default cleanup interval is changed to 1s.
  • Updated Sinker documentation. Starting from DataKit 2.0.0, non-point write APIs generally do not require __dataway_api rules.

Observability

  • Added Kodo queue metrics: dataway_kodo_queue_wait_seconds, dataway_kodo_queue_depth, dataway_kodo_queue_bytes, dataway_kodo_queue_enqueued_total, dataway_kodo_queue_full_total, and dataway_kodo_queue_dispatch_total.
  • Added invalid token negative cache metrics: dataway_token_negative_cache_added_total and dataway_token_negative_cache_blocked_total.
  • Added aggregation/tail sampling metrics: dataway_http_api_body_size_bytes_total, dataway_http_aggr_point_total, dataway_http_tail_sampling_trace_total, dataway_http_tail_sampling_span_total, and dataway_http_tail_sampling_packet_send_total.
  • Added disk cache lock contention metrics: diskcache_lock_wait_seconds and diskcache_lock_contention_total.
  • Fixed the sinked label of latency metrics in Sinker scenarios.

1.14.0 (2026/04/17)

  • Added a Session Replay image processing endpoint.

1.13.0 (2026/04/10)

  • Added support for bypass aggregation and sampling in DataWay (#67).

1.12.3 (2026/03/26)

  • Added v2 Sinker header format to resolve HTTP standard violations caused by abnormal values in routing headers (#68).

1.12.2 (2026/03/17)

  • Updated the SSL certificates in the DataWay image to avoid failure to recognize Openaway SSL certificates in cascading scenarios (#69).

1.12.1 (2025/12/25)

  • Fixed an issue in version 1.12.0 where Sinker routing could not copy requests (#66).

1.12.0 (2025/12/19)

New Features

  • Optimized the Sinker caching mechanism (#34).

    • The new sinker caching mechanism caches by a 16-byte request hash signature instead of caching the complete request signature. It uses dual hashing to avoid hash collisions as much as possible; the theoretical collision probability is n/2^128, where n is the number of cached keys.
    • Added TTL and capacity mechanisms to the Sinker cache to further limit memory usage: TTL cleans up inactive cache entries, and capacity ensures that the number of keys in the cache does not exceed the specified limit. The TTL and capacity limits further reduce the hash collision probability described above.
  • Added configuration to optimize DataWay HTTP header responses and reduce DataWay public network traffic (#65).


1.11.2 (2025/11/25)

  • Optimized DataWay forwarding of Kodo HTTP requests (#64).

1.11.1 (2025/11/15)

  • Fixed invalid __internal__ token issue (#63).

1.11.0 (2025/11/12)

New Features

  • Upgraded the DataWay base image to Ubuntu 22.04 (#62).
  • DataWay supports blocking invalid traffic (#61).

Bug Fixes

  • Fixed an issue where Session Replay and profile routing could not copy data to different Workspaces (#59).

1.10.0 (2025/09/03)

New Features

  • Added Langfuse data ingestion (#57).

1.9.0 (2025/04/25)

New Features

  • Added X-Pkg-Id to data upload requests in DataWay for data tracing (#53).
  • Improved application logs; more error codes returned by Kodo are passed through to DataKit (#52).
  • Added environment variables for API self-pull (#55).
  • Updated disk cache code to diagnose invalid argument errors that occur in some cases (#56).

1.8.0 (2025/02/19)

New Features

  • When Sinker mode is enabled, DataWay supports processing data reporting requests sent by clients (such as DataKit/Function) at the same time (#50).

1.7.0 (2024/12/18)

New Features

  • Added the AWS Firehose ingestion entry point (#47).
  • Added support for passing the token via X-Token (#43).

Feature Optimization

  • Adjusted default parameters for internal recycling to optimize memory usage (#49).
  • Optimized the default YAML in Kubernetes deployment and adjusted its cache disk mount configuration (#48).
  • Adjusted and added some internal metrics and updated the DataWay default dashboard.

1.6.2 (2024/12/03)

Feature Optimization

  • Added a connection configuration entry point at the HTTP layer (#46).

1.6.1 (2024/11/19)

Feature Optimization

  • Added an option to disable the 404 page (#44).

1.6.0 (2024/09/19)

Feature Optimization

  • Optimized environment variable configuration related to disk cache.
  • Exposed more self metrics and optimized memory usage at the HTTP forwarding layer (#39).
  • DataWay supports configuring its own SNI (#42).
  • Added a time synchronization endpoint to help DataKit obtain a more accurate Unix timestamp (#40).
  • Sinker:
    • Filter conditions support nil checks, i.e., determining whether a specific field exists (#41).
    • Added default rule settings, so requests that do not match existing routing rules are routed to the Workspace corresponding to the default rule (#30).

1.5.0 (2024/07/05)

Feature Optimization

  • Optimized the installation script (#33).

Compatibility Adjustment

  • The new version removes direct Sinker configuration support in Host installation mode. This feature will be supported later through a new approach.

1.4.1 (2024/06/19)

  • Fixed missing DW_ENABLE_TLS in Kubernetes installation mode.
  • Changed the build image address.

1.4.0 (2024/05/15)

  • Added DataKit metering endpoint (#29).
  • Fixed potential data loss in cache (#31).
  • Added support for direct HTTP TLS certificate configuration on the DataWay side (#32).
  • Exposed more metrics.

1.3.9 (2024/03/28)

  • When Sink drops requests, added HTTP 406 Not Acceptable error responses to facilitate troubleshooting (#82).

1.3.8 (2024/01/24)

  • Exposed more metrics.
  • Significantly improved Sinker rule matching performance (#26).

1.3.7 (2023/12/14)

  • Exposed more metrics.
  • Optimized the disk cache cleanup policy based on the principle of consuming as soon as possible:
    • Avoid actively dropping data due to storage overrun.
    • Send cached data as early as possible.
  • Added observability for dropping oversized requests (metrics/logs).
  • Adjusted DataWay's own Pod YAML and improved its metric and log collection.

1.3.6 (2023/10/24)


1.3.5 (2023/09/13)


1.3.4 (2023/09/01)

  • Upgraded Golang to 1.9.
  • Fixed the no-data issue with inner token.
  • Adjusted the following default configurations:
    • Default API rate limit increased to 100K.
    • Default maximum Body increased to 64MB.
    • Added HTTP timeout setting (DataWay to central service), default 30s.
    • Added configuration to ignore TLS 509 errors.

1.3.3 (2023/09/01)

  • Fixed curl issue in image packaging.

1.3.2 (2023/08/30)

  • Added the curl command to the image to facilitate viewing self metrics.

1.3.1 (2023/08/28)

  • Added Sinker functionality for URLs.

1.3.0 (2023/08/24)

  • Refactored the entire DataWay implementation and added Sinker functionality.

1.2.12 (2023/08/28)

  • Refactored the CI release process and implemented dual publishing of the install script.

1.2.8 (2022/09/18)

  • /v1/wirte endpoint updates:

    • Removed redundant /v1/write/xxx API and added /v1/write/:category endpoint.
    • Write data no longer performs line protocol decoding, but Body is still read for signing.
  • Disk cache:

    • When sending to the central service fails during cache cleanup, added a consumption pause policy so that the next cache is not cleaned until the current cached request is sent successfully.
    • Updated the pbcache proto file.

Feedback

Is this page helpful?