Prometheus

Release date 2026/09/15

Bugfix

• Fixed kong_ai_llm_tokens_total never reporting token counts for non-text-generation AI categories (image, video, realtime, and audio transcription generation), because it only read the legacy prompt_tokens/prompt_tokens_details analytics log fields instead of falling back to the input_tokens/input_tokens_details fields those categories are logged under.

• A histogram’s series (its buckets, _count and _sum) is now read as one atomic group during a scrape. Previously the exporter yielded and flushed output between individual samples, so an observation landing mid-scrape could bump _count/_sum after the buckets were already read, producing an internally inconsistent histogram in a single response.

• Fixed an issue where kong_nginx_requests_total and kong_nginx_connections_total duplicated the http values under the stream subsystem. They are now exported once under http.

• Histogram series are now emitted grouped by label set (all buckets in ascending le order, then _count, then _sum) so the output follows the OpenMetrics interleaving rule. Previously samples were grouped by suffix, which caused strict consumers such as Google Managed Prometheus to drop points.

• Histograms now expose every bucket (including those with a value of 0) starting from the first observation of a given label set. Previously only buckets greater than or equal to the observed value were created, so lower buckets appeared lazily on later scrapes.

Release date 2026/08/24

Bugfix

• A histogram’s series (its buckets, _count and _sum) is now read as one atomic group during a scrape. Previously the exporter yielded and flushed output between individual samples, so an observation landing mid-scrape could bump _count/_sum after the buckets were already read, producing an internally inconsistent histogram in a single response.

• Fixed an issue where kong_nginx_requests_total and kong_nginx_connections_total duplicated the http values under the stream subsystem. They are now exported once under http.

• Histograms now expose every bucket (including those with a value of 0) starting from the first observation of a given label set. Previously only buckets greater than or equal to the observed value were created, so lower buckets appeared lazily on later scrapes.

Release date 2026/08/20

Bugfix

• Fixed kong_ai_llm_tokens_total never reporting token counts for non-text-generation AI categories (image, video, realtime, and audio transcription generation), because it only read the legacy prompt_tokens/prompt_tokens_details analytics log fields instead of falling back to the input_tokens/input_tokens_details fields those categories are logged under.

Release date 2026/07/21

Bugfix

• Histogram series are now emitted grouped by label set (all buckets in ascending le order, then _count, then _sum) so the output follows the OpenMetrics interleaving rule. Previously samples were grouped by suffix, which caused strict consumers such as Google Managed Prometheus to drop points.

Release date 2026/08/26

Bugfix

• Fixed an issue where kong_nginx_requests_total and kong_nginx_connections_total duplicated the http values under the stream subsystem. They are now exported once under http.

Release date 2026/08/19

Bugfix

• Fixed kong_ai_llm_tokens_total never reporting token counts for non-text-generation AI categories (image, video, realtime, and audio transcription generation), because it only read the legacy prompt_tokens/prompt_tokens_details analytics log fields instead of falling back to the input_tokens/input_tokens_details fields those categories are logged under.

Release date 2026/07/14

Bugfix

• A histogram’s series (its buckets, _count and _sum) is now read as one atomic group during a scrape. Previously, the exporter yielded and flushed output between individual samples, so an observation landing mid-scrape could bump _count/_sum after the buckets were already read, producing an internally inconsistent histogram in a single response.

• Histogram series are now emitted grouped by label set (all buckets in ascending le order, then _count, then _sum) so the output follows the OpenMetrics interleaving rule. Previously, samples were grouped by suffix, which caused strict consumers such as Google Managed Prometheus to drop points.

• Histograms now expose every bucket (including those with a value of 0) starting from the first observation of a given label set. Previously, only buckets greater than or equal to the observed value were created, so lower buckets appeared lazily on later scrapes.

Release date 2025/12/18

Feature

• Added a new label type for http_requests_total and stream_sessions_total metrics providing more specific information.

Bugfix

• Fixed an issue where the Prometheus plugin would not return DB connections to the connection pool, potentially leading to exhaustion of available connections under high load.

Release date 2026/09/02

Bugfix

• Fixed an issue where kong_nginx_requests_total and kong_nginx_connections_total duplicated the http values under the stream subsystem. They are now exported once under http.

• Histogram series are now emitted grouped by label set (all buckets in ascending le order, then _count, then _sum) so the output follows the OpenMetrics interleaving rule. Previously samples were grouped by suffix, which caused strict consumers such as Google Managed Prometheus to drop points.

Release date 2025/10/01

Bugfix

• Fixed an issue where WebSocket wasn’t recorded in the Prometheus metrics.

• Fixed an issue where Prometheus metrics would show empty consumer identifiers when consumers didn’t have a username. The plugin now falls back to using the consumer’s custom_id when the username is not available, ensuring consumer identification in metrics.

Performance

• Optimized fetching of the kong_db_entities_total metric by using internal counters instead of the select count(*) query through the database. If this metric is found to be of sync with the real row counts, use the command kong migrations reinitialize-workspace-entity-counters to re-sync the counts.

Release date 2025/09/05

Bugfix

• Fixed an issue where Prometheus metrics would show empty consumer identifiers when consumers don’t have a username. The plugin now falls back to using the consumer’s custom_id when username is not available, ensuring consumer identification in metrics.

Performance

• Optimized fetching of the kong_db_entities_total metric by using internal counters instead of select count(*) query through the database. If this metric is found to be get out of sync with the real row counts, please use the command kong migrations reinitialize-workspace-entity-counters to re-sync the counts.

Release date 2025/07/03

Feature

• added consumer label for ai metrics and added metrics for ai-rag-injector plugin.

Bugfix

• Fixed an issue where AI latency histogram might miss some calculation.

• Fixed an issue where the control plane failed to expose the db_entities_total metric.

• Fixed an issue where the metric data_plane_config_hash might not work correctly for incremental sync.

Release date 2026/08/31

Bugfix

• Fixed an issue where kong_nginx_requests_total and kong_nginx_connections_total duplicated the http values under the stream subsystem. They are now exported once under http.

• Histogram series are now emitted grouped by label set (all buckets in ascending le order, then _count, then _sum) so the output follows the OpenMetrics interleaving rule. Previously samples were grouped by suffix, which caused strict consumers such as Google Managed Prometheus to drop points.

Release date 2025/10/10

Bugfix

• Fixed an issue where Prometheus metrics would show empty consumer identifiers when consumers don’t have a username. The plugin now falls back to using the consumer’s custom_id when username is not available, ensuring consumer identification in metrics.

Release date 2025/08/22

Performance

• Optimized fetching of the kong_db_entities_total metric by using internal counters instead of select count(*) query through the database. If this metric is found to be get out of sync with the real row counts, please use the command kong migrations reinitialize-workspace-entity-counters to re-sync the counts.

Release date 2025/07/06

Bugfix

• Fixed an issue where AI latency histogram might miss some calculation.

Release date 2025/04/15

Bugfix

• Fixed an issue where the metric data_plane_config_hash might not work correctly for incremental sync.

Release date 2025/03/27

Feature

• Added gauge to expose connectivity state to the control plane.

• Added the capability to enable or disable exporting of Proxy-Wasm metrics.

Release date 2024/12/12

Feature

• Bumped KONG_LATENCY_BUCKETS bucket’s maximal capacity to 6000

• Added support for Proxy-Wasm metrics.

Release date 2024/11/04

Feature

• Bumped KONG_LATENCY_BUCKETS bucket’s maximal capacity to 6000

Release date 2024/09/11

Feature

• AI plugins: retrieved latency data and pushed it to logs and metrics.

Bugfix

• Fixed an issue where CP/DP compatibility check was missing for the new configuration field ai_metrics.

• Improved error logging when having inconsistent labels count.

Release date 2024/05/28

Feature

• Added workspace label to Prometheus plugin metrics.

Release date 2024/02/12

Bugfix

• Expose metrics for serviceless routes

Release date 2023/08/09

Feature

• has been optimized to reduce proxy latency impacts during scraping. #10949#11040#11065

Release date 2022/12/06

Bugfix

• Added the missing protocols field to the following plugin schemas: Prometheus (prometheus)

Release date 2022/09/09

Feature

• nginx_http_current_connections and nginx_stream_current_connections were merged into to nginx_hconnections_total (or nginx_current_connections?)

• request_count and consumer_status were merged into http_requests_total.

  If the per_consumer config is set to false, the consumer label will be empty. If the per_consumer config is true, the consumer label will be filled.

• Updated the Grafana dashboard that comes packaged with Kong

Breaking Change

• High cardinality metrics are now disabled by default.

• Decreased performance penalty to proxy traffic when collecting metrics.

• http_status to http_requests_total.

• latency to kong_request_latency_ms (HTTP), kong_upstream_latency_ms, kong_kong_latency_ms, and session_duration_ms (stream).

  Kong latency and upstream latency can operate at orders of different magnitudes. Separate these buckets to reduce memory overhead.

• kong_bandwidth to kong_bandwidth_bytes.

• Removed the following metric: http_consumer_status

• New metrics:

  • session_duration_ms: monitoring stream connections.
  • node_info: Single gauge set to 1 that outputs the node’s ID and Kong Gateway version.

• session_duration_ms: monitoring stream connections.

• node_info: Single gauge set to 1 that outputs the node’s ID and Kong Gateway version.

• http_requests_total has a new label, source. It can be set to exit, error, or service.

• All memory metrics have a new label: node_id.

• The following metric names were adjusted to add units to standardize where possible:

  • http_status to http_requests_total.
  • latency to kong_request_latency_ms (HTTP), kong_upstream_latency_ms, kong_kong_latency_ms, and session_duration_ms (stream).

  Kong latency and upstream latency can operate at orders of different magnitudes. Separate these buckets to reduce memory overhead.

  • kong_bandwidth to kong_bandwidth_bytes.
  • nginx_http_current_connections and nginx_stream_current_connections were merged into to nginx_connections_total.
  • request_count and consumer_status were merged into http_requests_total.

  If the per_consumer config is set to false, the consumer label will be empty. If the per_consumer config is true, the consumer label will be filled.

• nginx_http_current_connections and nginx_stream_current_connections were merged into to nginx_connections_total.

• request_count and consumer_status were merged into http_requests_total.

  If the per_consumer config is set to false, the consumer label will be empty. If the per_consumer config is true, the consumer label will be filled.

• Updated the Grafana dashboard that comes packaged with Kong#8712

• The plugin doesn’t export status codes, latencies, bandwidth and upstream health check metrics by default. They can still be turned on manually by setting status_code_metrics,lantency_metrics, bandwidth_metrics and upstream_health_metrics respectively.#9028

Release date 2023/11/28

Performance

• reduce upstream health iteration latency spike during scrape

Release date 2023/03/28

Bugfix

• Added the missing protocols field to the following plugin schemas: Prometheus (prometheus)

Release date 2023/01/06

Bugfix

• Provided options to reduce the plugin’s impact on performance. Added new kong.conf options to switch high cardinality metrics on or off: prometheus_plugin_status_code_metrics, prometheus_plugin_latency_metrics, prometheus_plugin_bandwidth_metrics, and prometheus_plugin_upstream_health_metrics.

Release date 2022/03/02

Feature

• Added three new metrics:

  • kong_db_entities_total (gauge): total number of entities in the database.
  • kong_db_entity_count_errors (counter): measures the number of errors encountered during the measurement of kong_db_entities_total.
  • kong_nginx_timers (gauge): total number of Nginx timers, in Running or Pending state. Tracks ngx.timer.running_count() andngx.timer.pending_count().#8387

• kong_db_entities_total (gauge): total number of entities in the database.

• kong_db_entity_count_errors (counter): measures the number of errors encountered during the measurement of kong_db_entities_total.

• kong_nginx_timers (gauge): total number of Nginx timers, in Running or Pending state. Tracks ngx.timer.running_count() andngx.timer.pending_count().#8387