Performance#

The analyze subcommand was added to cloud-init to help analyze cloud-init boot time performance. It is loosely based on systemd-analyze, where there are four subcommands:

blame
show
dump
boot

Usage#

The analyze command requires one of the four subcommands:

$ cloud-init analyze blame
$ cloud-init analyze show
$ cloud-init analyze dump
$ cloud-init analyze boot

Availability#

The analyze boot subcommand only works on operating systems that use systemd.

Subcommands#

Blame#

The blame subcommand matches systemd-analyze blame where it prints, in descending order, the units that took the longest time to run. This output is highly useful for examining where cloud-init is spending its time.

$ cloud-init analyze blame

Example output:

-- Boot Record 01 --
80300s (init-network/config-growpart)
64300s (init-network/config-resizefs)
62100s (init-network/config-ssh)
57300s (modules-config/config-grub_dpkg)
40300s (init-local/search-NoCloud)
38200s (init-network/config-users_groups)
19800s (modules-config/config-apt_configure)
03700s (modules-final/config-keys_to_console)
02100s (init-network/config-update_etc_hosts)
02100s (init-network/check-cache)
00800s (modules-final/config-ssh_authkey_fingerprints)
00800s (init-network/consume-vendor-data)
00600s (modules-config/config-timezone)
00500s (modules-final/config-final_message)
00400s (init-network/consume-user-data)
00400s (init-network/config-mounts)
00400s (init-network/config-disk_setup)
00400s (init-network/config-bootcmd)
00400s (init-network/activate-datasource)
00300s (init-network/config-update_hostname)
00300s (init-network/config-set_hostname)
00200s (modules-final/config-snappy)
00200s (init-network/config-rsyslog)
00200s (init-network/config-ca_certs)
00200s (init-local/check-cache)
00100s (modules-final/config-scripts_vendor)
00100s (modules-final/config-scripts_per_once)
00100s (modules-final/config-salt_minion)
00100s (modules-final/config-rightscale_userdata)
00100s (modules-final/config-phone_home)
00100s (modules-final/config-package_update_upgrade_install)
00100s (modules-final/config-fan)
00100s (modules-config/config-ubuntu_advantage)
00100s (modules-config/config-ssh_import_id)
00100s (modules-config/config-snap)
00100s (modules-config/config-set_passwords)
00100s (modules-config/config-runcmd)
00100s (modules-config/config-locale)
00100s (modules-config/config-byobu)
00100s (modules-config/config-apt_pipelining)
00100s (init-network/config-write_files)
00100s (init-network/config-seed_random)
00100s (init-network/config-migrator)
00000s (modules-final/config-ubuntu_drivers)
00000s (modules-final/config-scripts_user)
00000s (modules-final/config-scripts_per_instance)
00000s (modules-final/config-scripts_per_boot)
00000s (modules-final/config-puppet)
00000s (modules-final/config-power_state_change)
00000s (modules-final/config-mcollective)
00000s (modules-final/config-lxd)
00000s (modules-final/config-landscape)
00000s (modules-final/config-chef)
00000s (modules-config/config-snap_config)
00000s (modules-config/config-ntp)
00000s (modules-config/config-disable_ec2_metadata)
00000s (init-network/setup-datasource)

boot records analyzed

Show#

The show subcommand is similar to systemd-analyze critical-chain which prints a list of units, the time they started and how long they took. Cloud-init has five boot stages, and within each stage a number of modules may run depending on configuration. cloudinit-analyze show will, for each boot, print this information and a summary of the total time.

The following is an abbreviated example of the show subcommand:

$ cloud-init analyze show

Example output:

-- Boot Record 01 --
The total time elapsed since completing an event is printed after the "@" character.
The time the event takes is printed after the "+" character.

Starting stage: init-local
|``->no cache found @00.01700s +00.00200s
|`->found local data from DataSourceNoCloud @00.11000s +00.40300s
Finished stage: (init-local) 00.94200 seconds

Starting stage: init-network
|`->restored from cache with run check: DataSourceNoCloud [seed=/dev/sr0][dsmode=net] @04.79500s +00.02100s
|`->setting up datasource @04.88900s +00.00000s
|`->reading and applying user-data @04.90100s +00.00400s
|`->reading and applying vendor-data @04.90500s +00.00800s
|`->activating datasource @04.95200s +00.00400s
Finished stage: (init-network) 02.72100 seconds

Starting stage: modules-config
|`->config-snap ran successfully @15.43100s +00.00100s
...
|`->config-runcmd ran successfully @16.22300s +00.00100s
|`->config-byobu ran successfully @16.23400s +00.00100s
Finished stage: (modules-config) 00.83500 seconds

Starting stage: modules-final
|`->config-snappy ran successfully @16.87400s +00.00200s
|`->config-package_update_upgrade_install ran successfully @16.87600s +00.00100s
...
|`->config-final_message ran successfully @16.93700s +00.00500s
|`->config-power_state_change ran successfully @16.94300s +00.00000s
Finished stage: (modules-final) 00.10300 seconds

Total Time: 4.60100 seconds

1 boot records analyzed

If additional boot records are detected then they are printed out from oldest to newest.

Dump#

The dump subcommand simply dumps the cloud-init logs that the analyze module is performing its analysis on, and returns a list of dictionaries that can be consumed for other reporting needs. Each element in the list is a boot entry.

$ cloud-init analyze dump

Example output:

[
{
  "description": "starting search for local datasources",
  "event_type": "start",
  "name": "init-local",
  "origin": "cloudinit",
  "timestamp": 1567057578.037
},
{
  "description": "attempting to read from cache [check]",
  "event_type": "start",
  "name": "init-local/check-cache",
  "origin": "cloudinit",
  "timestamp": 1567057578.054
},
{
  "description": "no cache found",
  "event_type": "finish",
  "name": "init-local/check-cache",
  "origin": "cloudinit",
  "result": "SUCCESS",
  "timestamp": 1567057578.056
},
{
  "description": "searching for local data from DataSourceNoCloud",
  "event_type": "start",
  "name": "init-local/search-NoCloud",
  "origin": "cloudinit",
  "timestamp": 1567057578.147
},
{
  "description": "found local data from DataSourceNoCloud",
  "event_type": "finish",
  "name": "init-local/search-NoCloud",
  "origin": "cloudinit",
  "result": "SUCCESS",
  "timestamp": 1567057578.55
},
{
  "description": "searching for local datasources",
  "event_type": "finish",
  "name": "init-local",
  "origin": "cloudinit",
  "result": "SUCCESS",
  "timestamp": 1567057578.979
},
{
  "description": "searching for network datasources",
  "event_type": "start",
  "name": "init-network",
  "origin": "cloudinit",
  "timestamp": 1567057582.814
},
{
  "description": "attempting to read from cache [trust]",
  "event_type": "start",
  "name": "init-network/check-cache",
  "origin": "cloudinit",
  "timestamp": 1567057582.832
},
...
{
  "description": "config-power_state_change ran successfully",
  "event_type": "finish",
  "name": "modules-final/config-power_state_change",
  "origin": "cloudinit",
  "result": "SUCCESS",
  "timestamp": 1567057594.98
},
{
  "description": "running modules for final",
  "event_type": "finish",
  "name": "modules-final",
  "origin": "cloudinit",
  "result": "SUCCESS",
  "timestamp": 1567057594.982
}
]

Boot#

The boot subcommand prints out kernel-related timestamps that are not included in any of the cloud-init logs. There are three different timestamps that are presented to the user:

kernel start
kernel finish boot
cloud-init start

This was added for additional clarity into the boot process that cloud-init does not have control over, to aid in debugging performance issues related to cloud-init startup, and tracking regression.

$ cloud-init analyze boot

Example output:

-- Most Recent Boot Record --
    Kernel Started at: 2019-08-29 01:35:37.753790
    Kernel ended boot at: 2019-08-29 01:35:38.807407
    Kernel time to boot (seconds): 1.053617000579834
    Cloud-init activated by systemd at: 2019-08-29 01:35:43.992460
    Time between Kernel end boot and Cloud-init activation (seconds): 5.185053110122681
    Cloud-init start: 2019-08-29 08:35:45.867000
successful

Timestamp gathering#

The following boot-related timestamps are gathered on demand when cloud-init analyze boot runs:

Kernel startup gathered from system uptime

Kernel finishes initialization from systemd UserSpaceMonotonicTimestamp property

Cloud-init activation from the property InactiveExitTimestamp of the cloud-init local systemd unit

In order to gather the necessary timestamps using systemd, running the following command will gather the UserspaceTimestamp:

$ systemctl show -p UserspaceTimestampMonotonic

Example output:

UserspaceTimestampMonotonic=989279

The UserspaceTimestamp tracks when the init system starts, which is used as an indicator of the kernel finishing initialisation.

Running the following command will gather the InactiveExitTimestamp:

$ systemctl show cloud-init-local -p InactiveExitTimestampMonotonic

Example output:

InactiveExitTimestampMonotonic=4493126

The InactiveExitTimestamp tracks when a particular systemd unit transitions from the Inactive to Active state, which can be used to mark the beginning of systemd’s activation of cloud-init.

Currently this only works for distros that use systemd as the init process. We will be expanding support for other distros in the future and this document will be updated accordingly.

If systemd is not present on the system, dmesg is used to attempt to find an event that logs the beginning of the init system. However, with this method only the first two timestamps are able to be found; dmesg does not monitor userspace processes, so no cloud-init start timestamps are emitted – unlike when using systemd.