Cloudprober’s main task is to run probes. A probe executes something, usually against a set of targets, to verify that the systems are working as expected from consumers’ point of view. For example, an HTTP probe executes an HTTP request against a web server to verify that the web server is available. Cloudprober probes run repeatedly at a configured interval and export probe results as a set of metrics.
A probe is defined as a set of the following fields:
|type||Probe type, for example: HTTP, PING or UDP|
|name||Probe name. Each probe should have a unique name.|
|interval_msec||How often to run the probe (in milliseconds).|
|timeout_msec||Probe timeout (in milliseconds).|
|targets||Targets to run probe against.|
Each probe also has optional probe-type specific config, for example, relative_url for HTTP probes. All probe types export following metrics at a minimum:
|total||Total number of probes.|
|success||Number of successful probes. Deficit between total and success indicates failures.|
|latency||Cumulative probe latency. Latency can be configured to be a distribution (histogram) metric through a config option. By default it’s just a counter like total and success.|
Cloudprober has built-in support for the following probe types:
More probe types can be added through cloudprober extensions (to be documented).
Ping probe type implements a fast ping prober, that can probe hundreds of targets in parallel. Probe results are reported as number of packets sent (total), received (success) and round-trip time (latency). It supports raw sockets (requires root access) as well as datagram sockets for ICMP (doesn’t require root access).
ICMP datagram sockets are not enabled by default on most Linux systems. You can
enable them by running the following command:
sudo sysctl -w net.ipv4.ping_group_range="0 5000"
HTTP probe is be used to send HTTP(s) requests to a target and verify that a response is received. Apart from core probe metrics (total, success and latency), HTTP probes also export a map of response code counts. Requests are marked as failed if there is a timeout.
UDP probe sends a UDP packet to the configured targets. UDP probe (and all other probes that use ports) provides more coverage for the network elements on the data path as most packet forwarding elements use 5-tuple hashing and using a new source port for each probe ensures that we hit different network element each time.
DNS probe type is implemented in a similar way as other probes except for that it sends DNS requests to the target.
External probe type allows running arbitrary probes through cloudprober. For an external probe, actual probe logic resides in an external program; cloudprober only manages the execution of that program and provides a way to export that data through the standard channel.
External probe can be configured in two modes:
ONCE: In this mode, an external program is executed for each probe run. Exit status of the program determines the success or failure of the probe. External probe can optionally be configured to interpret external program’s output as metrics. This is a simple model but it doesn’t allow the external program to maintain state and multiple forks can be expensive depending on the frequency of the probes.
SERVER: In this mode, external program is expected to run in server mode. Cloudprober automatically starts the external program if it’s not running at the time of the probe execution. Cloudprober and external probe process communicate with each other over stdin/stdout using protobuf messages defined in server.proto.