Overload manager

The overload manager is configured in the Bootstrap overload_manager field.

An example configuration of the overload manager is shown below. It shows a configuration to disable HTTP/1.x keepalive when heap memory usage reaches 95% and to stop accepting requests when heap memory usage reaches 99%.

refresh_interval:
  seconds: 0
  nanos: 250000000
resource_monitors:
  - name: "envoy.resource_monitors.fixed_heap"
    typed_config:
      "@type": type.googleapis.com/envoy.extensions.resource_monitors.fixed_heap.v3.FixedHeapConfig
      max_heap_size_bytes: 2147483648
actions:
  - name: "envoy.overload_actions.disable_http_keepalive"
    triggers:
      - name: "envoy.resource_monitors.fixed_heap"
        threshold:
          value: 0.95
  - name: "envoy.overload_actions.stop_accepting_requests"
    triggers:
      - name: "envoy.resource_monitors.fixed_heap"
        threshold:
          value: 0.99

Resource monitors

The overload manager uses Envoy’s extension framework for defining resource monitors. Envoy’s builtin resource monitors are listed here.

Triggers

Triggers connect resource monitors to actions. There are two types of triggers supported:

Overload actions

The following overload actions are supported:

Reducing timeouts

The envoy.overload_actions.reduce_timeouts overload action will reduce the amount of time Envoy will spend waiting for some interactions to finish in response to resource pressure. The amount of reduction can be configured per timeout type by specifying the minimum timer value to use when the triggering resource monitor detects saturation. The minimum value for each timeout can be specified either by providing a scale factor to apply to the configured maximum, or as a concrete duration value.

As an example, here is a single overload action entry that enables timeout reduction:

name: "envoy.overload_actions.reduce_timeouts"
triggers:
  - name: "envoy.resource_monitors.fixed_heap"
    scaled:
      scaling_threshold: 0.85
      saturation_threshold: 0.95
typed_config:
  "@type": type.googleapis.com/envoy.config.overload.v3.ScaleTimersOverloadActionConfig
  timer_scale_factors:
    - timer: HTTP_DOWNSTREAM_CONNECTION_IDLE
      min_timeout: 2s

It configures the overload manager to change the amount of time that HTTP connections are allowed to remain idle before being closed in response to heap size. When the heap usage is less than 85%, idle connections will time out at their usual time, which is configured through HttpConnectionManager.common_http_protocol_options.idle_timeout. When the heap usage is at or above 95%, idle connections will be closed after the specified min_timeout, here 2 seconds. If the heap usage is between 85% and 95%, the idle connection timeout will vary between those two based on the formula for the scaled trigger So if RouteAction.idle_timeout = 600 seconds and heap usage is at 92%, idle connections will time out after \(2s + (600s - 2s) \cdot (95\% - 92\%) / (95\% - 85\%) = 181.4s\).

Note in the example that the minimum idle time is specified as an absolute duration. If, instead, min_timeout: 2s were to be replaced with min_scale: { value: 10 }, the minimum timer value would be computed based on the maximum (specified elsewhere). So if idle_timeout is again 600 seconds, then the minimum timer value would be \(10\% \cdot 600s = 60s\).

Limiting Active Connections

Currently, the only supported way to limit the total number of active connections allowed across all listeners is via specifying an integer through the runtime key overload.global_downstream_max_connections. The connection limit is recommended to be less than half of the system’s file descriptor limit, to account for upstream connections, files, and other usage of file descriptors. If the value is unspecified, there is no global limit on the number of active downstream connections and Envoy will emit a warning indicating this at startup. To disable the warning without setting a limit on the number of active downstream connections, the runtime value may be set to a very large limit (~2e9). Listeners can opt out of this global connection limit by setting Listener.ignore_global_conn_limit to true. Similarly, you can opt out the admin listener by setting Admin.ignore_global_conn_limit. You may want to opt out a listener to be able to probe Envoy or collect stats while it is otherwise at its connection limit. Note that connections to listeners that opt out are still tracked and count towards the global limit.

If it is desired to only limit the number of downstream connections for a particular listener, per-listener limits can be set via the listener configuration.

One may simultaneously specify both per-listener and global downstream connection limits and the conditions will be enforced independently. For instance, if it is known that a particular listener should have a smaller number of open connections than others, one may specify a smaller connection limit for that specific listener and allow the global limit to enforce resource utilization among all listeners.

An example configuration can be found in the edge best practices document.

Reset Streams

Warning

Resetting streams via an overload action currently only works with HTTP2.

The envoy.overload_actions.reset_high_memory_stream overload action will reset expensive streams. This requires minimum_account_to_track_power_of_two to be configured via buffer_factory_config. To understand the memory class scheme in detail see minimum_account_to_track_power_of_two

As an example, here is a partial Overload Manager configuration with minimum threshold for tracking and a single overload action entry that resets streams:

buffer_factory_config:
  minimum_account_to_track_power_of_two: 20
actions:
  name: "envoy.overload_actions.reset_high_memory_stream"
  triggers:
    - name: "envoy.resource_monitors.fixed_heap"
      scaled:
        scaling_threshold: 0.85
        saturation_threshold: 0.95
...

We will only track streams using >= \(2^{minimum\_account\_to\_track\_power\_of\_two}\) worth of allocated memory in buffers. In this case, by setting the minimum_account_to_track_power_of_two to 20 we will track streams using >= 1MiB since \(2^{20}\) is 1MiB. Streams using >= 1MiB will be classified into 8 power of two sized buckets. Currently, the number of buckets is hardcoded to 8. For this example, the buckets are as follows:

The above configuration also configures the overload manager to reset our tracked streams based on heap usage as a trigger. When the heap usage is less than 85%, no streams will be reset. When heap usage is at or above 85%, we start to reset buckets according to the strategy described below. When the heap usage is at 95% all streams using >= 1MiB memory are eligible for reset. This overload action will reset up to 50 streams (this is a hardcoded limit) per worker everytime the action is invoked. This is both to reduce the amount of streams that end up getting reset and to prevent the worker thread from locking up and triggering the Watchdog system.

Given that there are only 8 buckets, we partition the space with a gradation of \(gradation = (saturation\_threshold - scaling\_threshold)/8\). Hence at 85% heap usage we reset streams in the last bucket e.g. those using >= 128MiB. At \(85% + 1 * gradation\) heap usage we reset streams in the last two buckets e.g. those using >= 64MiB, prioritizing the streams in the last bucket since there’s a hard limit on the number of streams we can reset per invokation. At \(85% + 2 * gradation\) heap usage we reset streams in the last three buckets e.g. those using >= 32MiB. And so forth as the heap usage is higher.

It’s expected that the first few gradations shouldn’t trigger anything, unless there’s something seriously wrong e.g. in this example streams using >= 128MiB in buffers.

Statistics

Each configured resource monitor has a statistics tree rooted at overload.<name>. with the following statistics:

Each configured overload action has a statistics tree rooted at overload.<name>. with the following statistics: