Configure the internal queue

Metricbeat uses an internal queue to store events before publishing them. The queue is responsible for buffering and combining events into batches that can be consumed by the outputs. The outputs will use bulk operations to send a batch of events in one transaction.

You can configure the type and behavior of the internal queue by setting options in the queue section of the metricbeat.yml config file. Only one queue type can be configured.

This sample configuration sets the memory queue to buffer up to 4096 events:

queue.mem:
  events: 4096

The memory queue keeps all events in memory.

If no flush interval and no number of events to flush is configured, all events published to this queue will be directly consumed by the outputs. To enforce spooling in the queue, set the flush.min_events and flush.timeout options.

By default flush.min_events is set to 2048 and flush.timeout is set to 1s.

The output’s bulk_max_size setting limits the number of events being processed at once.

The memory queue waits for the output to acknowledge or drop events. If the queue is full, no new events can be inserted into the memory queue. Only after the signal from the output will the queue free up space for more events to be accepted.

This sample configuration forwards events to the output if 512 events are available or the oldest available event has been waiting for 5s in the queue:

queue.mem:
  events: 4096
  flush.min_events: 512
  flush.timeout: 5s

You can specify the following options in the queue.mem section of the metricbeat.yml config file:

Number of events the queue can store.

The default value is 4096 events.

Minimum number of events required for publishing. If this value is set to 0, the output can start publishing events without additional waiting times. Otherwise the output has to wait for more events to become available.

The default value is 2048.

Maximum wait time for flush.min_events to be fulfilled. If set to 0s, events will be immediately available for consumption.

The default value is 1s.

The disk queue stores pending events on the disk rather than main memory. This allows Beats to queue a larger number of events than is possible with the memory queue, and to save events when a Beat or device is restarted. This increased reliability comes with a performance tradeoff, as every incoming event must be written and read from the device’s disk. However, for setups where the disk is not the main bottleneck, the disk queue gives a simple and relatively low-overhead way to add a layer of robustness to incoming event data.

The disk queue is expected to replace the file spool in a future release.

To enable the disk queue with default settings, specify a maximum size:

queue.disk:
  max_size: 10GB

The queue will use up to the specified maximum size on disk. It will only use as much space as required. For example, if the queue is only storing 1GB of events, then it will only occupy 1GB on disk no matter how high the maximum is. Queue data is deleted from disk after it has been successfully sent to the output.

You can specify the following options in the queue.disk section of the metricbeat.yml config file:

The path to the directory where the disk queue should store its data files. The directory is created on startup if it doesn’t exist.

The default value is "${path.data}/diskqueue".

The maximum size the queue should use on disk. Events that exceed this maximum will either pause their input or be discarded, depending on the input’s configuration.

A value of 0 means that no maximum size is enforced, and the queue can grow up to the amount of free space on the disk. This value should be used with caution, as completely filling a system’s main disk can make it inoperable. It is best to use this setting only with a dedicated data or backup partition that will not interfere with Metricbeat or the rest of the host system.

The default value is 10GB.

Data added to the queue is stored in segment files. Each segment contains some number of events waiting to be sent to the outputs, and is deleted when all its events are sent. By default, segment size is limited to 1/10 of the maximum queue size. Using a smaller size means that the queue will use more data files, but they will be deleted more quickly after use. Using a larger size means some data will take longer to delete, but the queue will use fewer auxiliary files. It is usually fine to leave this value unchanged.

The default value is max_size / 10.

The number of events that should be read from disk into memory while waiting for an output to request them. If you find outputs are slowing down because they can’t read as many events at a time, adjusting this setting upward may help, at the cost of higher memory usage.

The default value is 512.

The number of events the queue should accept and store in memory while waiting for them to be written to disk. If you find the queue’s memory use is too high because events are waiting too long to be written to disk, adjusting this setting downward may help, at the cost of reduced event throughput. On the other hand, if inputs are waiting or discarding events because they are being produced faster than the disk can handle, adjusting this setting upward may help, at the cost of higher memory usage.

The default value is 2048.

Some disk errors may block operation of the queue, for example a permission error writing to the data directory, or a disk full error while writing an event. In this case, the queue reports the error and retries after pausing for the time specified in retry_interval.

The default value is 1s (one second).

When there are multiple consecutive errors writing to the disk, the queue increases the retry interval by factors of 2 up to a maximum of max_retry_interval. Increase this value if you are concerned about logging too many errors or overloading the host system if the target disk becomes unavailable for an extended time.

The default value is 30s (thirty seconds).

The file spool queue stores all events in an on disk ring buffer. The spool has a write buffer, which new events are written to. Events written to the spool are forwarded to the outputs, only after the write buffer has been flushed successfully.

The spool waits for the output to acknowledge or drop events. If the spool is full, no new events can be inserted. The spool will block. Space is freed only after a signal from the output has been received.

On disk, the spool divides a file into pages. The file.page_size setting configures the file’s page size at file creation time. The optimal page size depends on the effective block size, used by the underlying file system.

This sample configuration enables the spool with all default settings (See Configuration options for defaults) and the default file path:

queue.spool: ~

This sample configuration creates a spool of 512MiB, with 16KiB pages. The write buffer is flushed if 10MiB of contents, or 1024 events have been written. If the oldest available event has been waiting for 5s in the write buffer, the buffer will be flushed as well:

queue.spool:
  file:
    path: "${path.data}/spool.dat"
    size: 512MiB
    page_size: 16KiB
  write:
    buffer_size: 10MiB
    flush.timeout: 5s
    flush.events: 1024

You can specify the following options in the queue.spool section of the metricbeat.yml config file:

The spool file path. The file is created on startup, if it does not exist.

The default value is "${path.data}/spool.dat".

The file permissions. The permissions are applied when the file is created. In case the file already exists, the file permissions are compared with file.permissions. The spool file is not opened if the actual file permissions are more permissive then configured.

The default value is 0600.

Spool file size.

The default value is 100 MiB.

The file’s page size.

The spool file is split into pages of page_size. All I/O operations operate on complete pages.

The default value is 4096 (4KiB).

If prealloc is set to true, truncate is used to reserve the space up to file.size. This setting is only used when the file is created.

The file will dynamically grow, if prealloc is set to false. The spool blocks, if prealloc is false and the system is out of disk space.

The default value is true.

The write buffer size. The write buffer is flushed, once the buffer size is exceeded.

Very big events are allowed to be bigger then the configured buffer size. But the write buffer will be flushed right after the event has been serialized.

The default value is 1MiB.

The event encoding used for serialized events. Valid values are json and cbor.

The default value is cbor.

Maximum wait time of the oldest event in the write buffer. If set to 0, the write buffer will only be flushed once write.flush.events or write.buffer_size is fulfilled.

The default value is 1s.

Number of buffered events. The write buffer is flushed once the limit is reached.

The default value is 16384.

The spool reader tries to read up to the output’s bulk_max_size events at once.

If read.flush.timeout is set to 0s, all available events are forwarded immediately to the output.

If read.flush.timeout is set to a value bigger then 0s, the spool will wait for more events to be flushed. Events are forwarded to the output if bulk_max_size events have been read or the oldest read event has been waiting for the configured duration.

The default value is 0s.