fleet/cmd/osquery-perf

Osquery Server Performance Tester

This is a tool to generate realistic traffic to an osquery management server (primarily, Fleet). With this tool, many thousands of hosts can be simulated from a single host.

Requirements

The only requirement for running this tool is a working installation of Go.

Usage

Typically go run is used.

You can use --help to view the available configuration:

go run agent.go --help

The tool should be invoked with the appropriate enroll secret. A typical invocation looks like:

go run agent.go --enroll_secret hgh4hk3434l2jjf

When starting many hosts, it is a good idea to extend the intervals, and also the period over which the hosts are started:

go run agent.go --enroll_secret hgh4hk3434l2jjf --host_count 5000 --start_period 5m --query_interval 60s --config_interval 5m

This will start 5,000 hosts over a period of 5 minutes. Each host will check in for live queries at a 1 minute interval, and for configuration at a 5 minute interval. Starting over a 5 minute period ensures that the configuration requests are spread evenly over the 5 minute interval.

It can be useful to start the "same" hosts. This can be achieved with the --seed parameter:

go run agent.go --enroll_secret hgh4hk3434l2jjf --seed 0

By using the same seed, along with other values, we usually get hosts that look the same to the server. This is not guaranteed, but it is a useful technique.

By default, all hosts will simulate macOS hosts (specifically, macOS 10.14). To simulate hosts using other operating systems, use the --os_templates flag. This flag takes a comma-separated list of host template names and will start hosts by alternating in the list of OS templates when multiple templates are specified. For example:

go run agent.go --enroll_secret hgh4hk3434l2jjf --os_templates ubuntu_22.04,windows_11 --host_count 6

would start 3 Ubuntu hosts and 3 Windows hosts.

Supported Linux templates: ubuntu_22.04, rhel_8, rhel_9, rhel_10. RHEL templates report platform=rhel, RPM-style kernels (e.g., kernel-core 5.14.0-503.26.1.el9_5), and (when --software_db_path points at a populated database) additional RPM packages from the software library. See the os_templates flag description in go run agent.go --help for the full list of supported template names.

The software database (cmd/osquery-perf/software-library/software.db) is optional — macOS, Windows, and Ubuntu have embedded fallback fixtures, and RHEL kernels are embedded too. The DB only adds non-kernel software variety. If the DB isn't present at --software_db_path, osquery-perf logs a warning and falls back to the embedded fixtures.

Controlling Agent Behavior From the Fleet UI

Specify Query Results

Using the naming convention MyQuery_10 (name separated by _number) will instruct agents to return 10 rows for that query

Control policy pass/fail per policy

In the Policy SQL:

• select 1 will instruct agents to send back only passing responses
• select 0 will instruct agents to send back only failing responses

Running Locally (Development Environment)

First, ensure your Fleet local development environment is up and running. Refer to Building Fleet for details. Once this is done:

• navigate to the Hosts tab of your Fleet web interface (typically, this would be at https://localhost:8080/hosts/manage).
• click on "Manage enroll secret" and copy the enroll secret.
• start the osquery-perf agent (from the root of the Fleet repository, it would be go run ./cmd/osquery-perf/agent.go --enroll_secret <paste-the-secret>).

Alternatively, you can retrieve the enroll secret from the command-line using fleetctl get enroll_secret (you may have to login to fleetctl first).

The agent will start. You can connect to MySQL to view changes made to the development database by the agent (e.g., at the terminal, with docker-compose exec mysql mysql -uroot -ptoor -Dfleet). Remember that frequency of the reported data depends on the configuration of the Fleet instance, so you may want to start it with shorter delays for some cases and enable debug logging (e.g., ./build/fleet serve --dev --logging_debug --osquery_detail_update_interval 1m).

Resource Limits

On many systems, trying to simulate a large number of hosts will result in hitting system resource limits (such as number of open file descriptors).

If you see errors such as dial tcp: lookup localhost: no such host or read: connection reset by peer, try increasing these limits.

macOS

Run the following command in the shell before running the Fleet server and before running agent.go (run it once in each shell):

ulimit -n 64000

Running with MDM

Set up MDM on your server. To extract the SCEP challenge, you can use the MDM asset extractor.

For your server, disable Apple push notifications since we will be using devices with fake UUIDs:

export FLEET_DEV_MDM_APPLE_DISABLE_PUSH=1

Example of running the agent with MDM. Note that enroll_secret is not needed for iPhone/iPad devices:

go run agent.go --os_templates ipad_13.18,iphone_14.6 --host_count 10 --mdm_scep_challenge 0d53306e-6d7a-9d14-a372-f9e53f9d62db

mdm_prob determines the probability of MDM enrollment for each host. The default is 0 (0%). You can set it to 1.0 to ensure all hosts enroll in MDM.

mdm_user_prob determines the probability of MDM user enrollment for each host. The default is 0 (0%). You can set it to 1.0 to ensure all hosts enroll in MDM user enrollment. This probability stacks with mdm_prob. So this probability is based on the hosts who end up MDM enrolling.

Apple Platform SSO (PSSO)

A subset of macOS MDM agents can additionally exercise Apple Platform SSO: device registration, password login (proxied through Fleet to your IdP), and the offline-unlock key request/exchange. This requires a server that has account provisioning configured (with a reachable ROPG IdP) and the PSSO configuration profile assigned to the enrolled hosts — the agent obtains its Fleet-signed registration token from the delivered profile, so nothing happens until that profile reconciles onto the host.

Each selected agent registers once (staggered across --mdm_psso_interval to avoid a thundering herd), does one login and one key request/exchange during setup, and then, on each interval, performs a login and/or a key request/exchange according to their probabilities — spread across the interval rather than on the tick boundary.

• --mdm_psso_prob: default 0, probability an MDM-enrolled macOS host also simulates PSSO [0, 1]
• --mdm_psso_client_id: IdP/extension client ID, must match the server's account provisioning config (PSSO is skipped when empty)
• --mdm_psso_username / --mdm_psso_password: IdP credentials used for logins (must be accepted by the IdP Fleet proxies to)
• --mdm_psso_interval: default 4h, window for staggering registrations and recurring logins/key operations
• --mdm_psso_login_prob: default 1.0, probability of a login during each interval after registration [0, 1]
• --mdm_psso_key_prob: default 0.1, probability of a key request/exchange during each interval after registration [0, 1]

go run agent.go --host_count 100 --mdm_prob 1.0 --mdm_scep_challenge <challenge> \
  --mdm_psso_prob 0.5 --mdm_psso_client_id <client-id> \
  --mdm_psso_username loadtest@example.com --mdm_psso_password <password> \
  --mdm_psso_interval 4h --mdm_psso_login_prob 1.0 --mdm_psso_key_prob 0.1

Synthetically reproducing MDM device protocol failures

NotNow'ing profiles

Currently only supported for macOS and InstallProfile commands

To force an osquery-perf agent to respond with NotNow once to an InstallProfile command, the payload has to contain NotNow anywhere in the profile. It will NotNow once, then acknowledge it on next check-in. To force a new NotNow response, you have to change the ProfileIdentifier.

Forcing a certain error code and failure for InstallApplication

Currently only supported for macOS.

To force a certain ErrorCode and failure for an InstallApplication command, the iTunesStoreID payload field has to have a value below 100_000. The agent will respond with a failure and the specified error code, which helps QA and repro logic scenarios on certain error codes.

Installing software

The agent can install software for "macos", "ubuntu", and "windows" OSs when running with orbit agent. The following options control the installation behavior:

• --software_installer_pre_install_fail_prob: default 0.05, select 1 always passes and select 0 always fails
• --software_installer_install_fail_prob: default 0.05, exit 0 always passes and exit 1 always fails
• --software_installer_post_install_fail_prob: default 0.05, exit 0 always passes and exit 1 always fails