1
0
mirror of https://gitlab.alpinelinux.org/alpine/cloud/tiny-cloud.git synced 2025-12-15 11:22:43 +03:00

Final Bits Before 3.0.0

This commit is contained in:
Jake Buchholz Göktürk 2023-05-29 22:39:36 +00:00
parent 82eb4e7d0c
commit 96e36c251d
6 changed files with 153 additions and 90 deletions

View File

@ -2,18 +2,46 @@
## 2023-05-XX - Tiny Cloud v3.0.0
### INIT SCRIPTS / PHASES
* Tiny Cloud has been reorganized into **four** init script phases...
* `boot` - "boot" runlevel, after `syslogd` is started, and before
networking is up.
* `early` - "default" runlevel, after networking is up, but *before*
user-data is made available. This phase is responsible for pulling in
user-data for *later* phases.
* `main` - "default" runlevel, after user-data is available. Most things
get done here.
* `final` - "default" runlevel, after all other init scripts have finished.
Typically user-data scripts are run here, and bootstrap status is marked
as complete.
* Tiny Cloud init functionality has been consolidated into **/sbin/tiny-cloud**
and init scripts should use `tiny-cloud <phase>` to indicate whether `early`,
`main`, or `final` actions should be taken. Additionally, it is now possible
for clouds to specify their own (or supercede the default) init functions
and/or change which init phase they are executed in.
and init scripts should use `tiny-cloud <phase>` to indicate whether `boot`,
`early`, `main`, or `final` actions should be taken.
* Use `tiny-cloud --setup` to add/update Tiny Cloud's init scripts into the
right runlevels. Currently only OpenRC is supported.
* The example OpenRC init scripts been updated and moved to **dist/openrc/**.
Use `tiny-cloud --setup` to add Tiny Cloud's init scripts into the right
runlevels
### NEW STUFF
* Tiny Cloud configuration has moved to **/etc/tiny-cloud.conf**.
* Tiny Cloud now supports the concept of user-data handlers, to support acting
on different payload content-types.
* Clouds and user-data handlers can specify their own (or supercede the default)
init functions and/or change the init phase in which they are executed. The
order of declaration is default --> cloud --> user-data (last one wins).
* We now have an experimental `alpine` installer "cloud", based on NoCloud,
and an associated `#alpine-config` user-data handler, which supports a subset
of `#cloud-config` features, plus some extensions. As this is new, we expect
it to experience some continued evolution.
* Also thanks to Nataniel Copa (@ncopa), Tiny Cloud now has unit tests.
### DEPRECATION
* `nvme-ebs-symlinks` has been _deprecated_ and disabled by default. The
**mdev-conf** package, as of v4.4 is now responsible for maintaining NVMe
@ -23,6 +51,10 @@
**/dev/sd** or **/dev/xvd** symlinks are created as indicated in NVMe device
metadata, *but NOT both*!
### MISCELLANEOUS
* Tiny Cloud configuration has moved to **/etc/tiny-cloud.conf**.
* `imds` now supports `@local-hostname` alias. For most clouds this is the
same as `@hostname`.

144
README.md
View File

@ -8,57 +8,64 @@ do just what is necessary with a small footprint and minimal dependencies.
A direct descendant of [tiny-ec2-bootstrap](
https://gitlab.alpinelinux.org/alpine/cloud/tiny-ec2-bootstrap), Tiny Cloud
works with multiple cloud providers. Currently, the following are supported:
* [AWS](https://aws.amazon.com): Amazon Web Services
* [Azure](https://azure.microsoft.com): Microsoft Azure
* [GCP](https://cloud.google.com): Google Cloud Platform
* [OCI](https://cloud.oracle.com): Oracle Cloud Infrastructure
* [AWS](https://aws.amazon.com) - Amazon Web Services
* [Azure](https://azure.microsoft.com) - Microsoft Azure
* [GCP](https://cloud.google.com) - Google Cloud Platform
* [OCI](https://cloud.oracle.com) - Oracle Cloud Infrastructure
* [NoCloud](
https://cloudinit.readthedocs.io/en/latest/reference/datasources/nocloud.html
):
cloud-init's NoCloud AWS-compatible user provided data source
https://cloudinit.readthedocs.io/en/latest/reference/datasources/nocloud.html
) - cloud-init's NoCloud AWS-compatible user provided data source
Tiny Cloud is also used for Alpine Linux's experimental "auto-install" feature.
## Features
The following actions will occur ***only once***, during the initial boot of an
instance:
* expand the root filesystem to use all available root device space, during the
**sysinit** runlevel
* set the instance's hostname from instance metadata
* install SSH keys from instance metadata to the cloud user account's
`authorized_keys` file (the user must already exist)
* save the instance user-data to a file, and if it's a script, execute it at
the end of the **default** runlevel
* expand the root filesystem to use all available root device space
* set up default network interfaces, if necessary
* enable `sshd`, if necessary
* save instance user-data to a file, decompress if necessary
* create default cloud user, if necessary
* set the instance's hostname from instance meta-data
* install SSH keys from instance meta-data to the cloud user account's
`authorized_keys` file
* if instance user-data is a script, execute it at the end of the **default**
runlevel
* mark the bootstrap of the instance as "complete"
Optional features, which may not be universally necessary:
* manage hotpluggable network interfaces
* sync IMDS-provided secondary IPv4 and IPv6 addresses network interfaces
* manage symlinks from NVMe block devices to `/dev/xvd` or `/dev/sd` devices
(i.e. AWS Nitro instances)
* manage hotpluggable virtual network interfaces
* sync IMDS-provided secondary IPv4 and IPv6 network configuration
Other cloud- and user-data-specific actions may also occur.
Also included is a handy `imds` client script for easy access to an instance's
IMDS data.
## Requirements
As Tiny Cloud is meant to be tiny, it has very few dependencies:
As Tiny Cloud is meant to be tiny, it has few dependencies:
* Busybox (`ash`, `wget`, etc.)
* `ifupdown-ng` (optional, for network management)
* `iproute2-minimal` (optional, for syncing IPv4/IPv6 from IMDS)
* `nvme-cli` (optional, for AWS nitro NVMe symlinks)
* `e2fsprogs-extra` (for `resize2fs`)
* `openssh-server`
* `partx`
* `resize2fs`
* `sfdisk`
* [`yx`](https://gitlab.com/tomalok/yx)
(optional, allows NoCloud to extract metadata from YAML files)
* [`yx`](https://gitlab.com/tomalok/yx) (for extracting data from YAML files)
Tiny Cloud has been developed specifically for use with the
Optional dependencies:
* `ifupdown-ng` (for network management)
* `iproute2-minimal` (for syncing IPv4/IPv6 from IMDS)
* `nvme-cli` (for AWS nitro NVMe symlinks)
_Tiny Cloud has been developed specifically for use with the
[Alpine Cloud Images](
https://gitlab.alpinelinux.org/alpine/cloud/alpine-cloud-images)
project, and as such, it is currently tailored for use with [Alpine Linux](
https://alpinelinux.org), the [OpenRC](https://github.com/OpenRC/openrc) init
system, and the `ext4` root filesystem. If you would like to see Tiny Cloud
supported on additional distributions, init systems, and/or filesystems, please
open an issue with your request -- or better yet, submit a merge request!
open an issue with your request -- or better yet, submit a merge request!_
## Installation
@ -73,13 +80,13 @@ dependencies for Tiny Cloud to support _`<cloud>`_.
Alternately, you can download a release tarball, and use `make` to install it.
Next, enable the three primary init scripts...
Next, set up the RC scripts...
```
rc-update add tiny-cloud-early sysinit
rc-update add tiny-cloud default
rc-update add tiny-cloud-final default
tiny-cloud --setup
```
That's it! On the next boot, Tiny Cloud will bootstrap the instance.
## Configuration
By default, Tiny Cloud expects configuration at `/etc/tiny-cloud.conf`,
@ -94,40 +101,50 @@ Current valid values are `aws`, `azure`, `gcp`, `oci`, and `nocloud`._
The first time an instance boots -- either freshly instantiated from an image,
or after installation on a pre-existing instance -- Tiny Cloud sets up the
instance in three phases...
instance in four phases...
### Boot Phase
This phase does not depend on the cloud provider's Instance Meta-Data Service
(IMDS), and does not require networking to be up. `mdev` hotplug modules are
installed (if any), default networking confinguration is set up, `sshd` is
enabled (but not started), and the root partition is expanded.
### Early Phase
The `tiny-cloud-early` init script does not depend on the cloud provider's
Instance MetaData Service (IMDS), and therefore does not have a dependency on
networking. During this "early" phase, the root filesystem is expanded, and
any necessary `mdev` rules for device hotplug are set up.
After networking is up, and the cloud provider's IMDS is available, this phase
is primarily responsible for retrieving the instance's User-Data for use by
later phases.
User-Data is stored at `/var/lib/cloud/user-data`, and will be decompressed, if
necessary. Currently supported compression algorithms are `gzip`, `bzip2`,
`unxz`, `lzma`, `lzop`, `lz4`, and `zstd`. _(Note that `lz4` and `zstd` are
not installed in Alpine by default, and would need to be added to the image.)_
### Main Phase
The main `tiny-cloud` init script *does* depend on the cloud provider's IMDS
data; it saves the instance's user data to `/var/lib/cloud/user-data`, sets up
instance's hostname, and installs the cloud user's SSH keys before `sshd`
starts.
When networking, IMDS, and User-Data are all availabile, this is the phase
takes care of the majority of bootstrapping actions that require them --
setting up the instanxe hostname, creating default cloud user, and installing
SSH keys for it.
If the user data is compressed, Tiny Cloud will decompress it. Currently
supported compression algorithms are `gzip`, `bzip2`, `unxz`, `lzma`, `lzop`,
`lz4`, and `zstd`. _(Note that `lz4` and `zstd` are not installed in Alpine
by default, and would need to be added to the image.)_
Additional main phase actions may be taken if there is a User-Data handler
defined for its content type, and those actions are associated with the main
phase.
### Final Phase
`tiny-cloud-final` should be the very last init script to run in the
**default** runlevel.
The very last thing to be run in the **default** runlevel this phase will
execute the saved User-Data, if it is a script starting with `#!`; its output
(combined STDOUT and STDER) and exit code are saved to `/var/log/user-data.log`
and `/var/log/user-data.exit`.
If the user data is a script starting with `#!/`, it will be executed; its
output (combined STDOUT and STDERR) and exit code are saved to
`/var/log/user-data.log` and `/var/log/user-data.exit`, respectively; the
directory is overrideable via the `TINY_CLOUD_LOGS` config setting.
Additional final phase actions may be taken if there is a User-Data handler
defined for its content type, and those actions are associated with the final
phase.
If all went well, the very last thing `tiny-cloud-final` does is touch
a `.bootstrap-complete` file into existence in `/var/lib/cloud` or another
directory specified by the `TINY_CLOUD_VAR` config setting.
The very last action to be taken is to mark the instance's bootstrap as
"complete", so that future reboots do not re-boostrap the instance.
### Skipping Init Actions
@ -142,17 +159,19 @@ no-op.
To force the init scripts to re-run on the next boot...
```
rm -f /var/lib/cloud/.bootstrap-complete
```
or
```
service tiny-cloud incomplete
tiny-cloud --bootstrap incomplete
```
If you're instantiating an instance in order to create a new cloud image
(using [Packer](https://packer.io), or some other means), you will need to
do this before creating the image to ensure that instances using the new
image will also run Tiny Cloud init scripts during their first boot.
To check the status of the Tiny Cloud bootstrap, use...
```
tiny-cloud --bootstrap status
```
...which will either respond with `complete` or `incomplete`
## Cloud Hotplug Modules
### `vnic_eth_hotplug`
@ -163,12 +182,3 @@ attached/detached from the instance.
An `ifupdown-ng` executor also syncs the interfaces' secondary IPv4 and IPV6
addresses associated with those VNICs, if the cloud's IMDS provides that
configuration data.
### `nvme_ebs_links`
EBS volumes are attached to AWS EC2 Nitro instances using the NVMe driver.
Unfortunately, the `/dev/nvme*` device names do not match the device name
assigned to the attached EBS volume. This hotplug module figures out what the assigned device name is, and sets up `/dev/xvd*` and `/dev/sd*` symlinks to
the right NVMe devices for EBS volumes and their partitions.
_(deprecated, see [CHANGELOG.md](CHANGELOG.md))_

15
TODO.md
View File

@ -1,10 +1,19 @@
# TODO
## SOON-ish
* Should the extra stuff that the `alpine` installer cloud does also apply to
the `nocloud` cloud? If so, move it there, and the installer is entirely
handled by the user-data handler.
* Package user-data handlers separately?
* `#cloud-config` user-data handler (support a useful subset)
## FUTURE
* cloud auto-detection
* `#cloud-config` user-data handler (support a useful subset)
* cloud auto-detection?
* `#tiny-config` user-data handler (should be simple-yet-flexible)

View File

@ -13,6 +13,7 @@ if [ "$1" = "-h" ] || [ "$1" = "--help" ]; then
Usage: imds [-h] { -e | +e | +n | +s | +t | @<alias> | <imds-path> } ...
-h : help
-e / +e : ignore / catch errors
-E / +E : hide / slow STDERR
+n / +s / +t : insert newline / space / tab
<alias> :-
hostname : instance hostname
@ -94,15 +95,17 @@ fi
### non-overrideable functions
imds() {
local cmd args key rv err=1
local cmd args key rv err=1 stderr=1
while [ -n "$1" ]; do
cmd=_imds
args=
key="$1"; shift
case $key in
# error handling
# error handling/output
-e) err=0; continue ;; # ignore
+e) err=1; continue ;; # return
-E) stderr=0; continue ;; # hide
+E) stderr=1; continue ;; # show
# TODO: retry/deadline
# output control
+n) printf "\n"; continue ;; # insert newline
@ -122,8 +125,13 @@ imds() {
*) args="$key" ;;
esac
# TODO: retry/deadline
"$cmd" $args
rv=$?
if [ $stderr = 1 ]; then
"$cmd" $args
rv=$?
else
"$cmd" $args 2>/dev/null
rv=$?
fi
[ $err -eq 0 ] && continue
[ $rv = "0" ] || return $rv
done

View File

@ -49,7 +49,7 @@ iface_ip6s() {
}
imds_ip4s() {
local ip4=$(imds "@nic:$IFACE,@ipv4")
local ip4=$(imds -E "@nic:$IFACE,@ipv4")
local ip4s=$(echo "$ip4" | tail +2) # secondary IPv4s
local ip4p ip4_cidr ip4_gw
@ -58,7 +58,7 @@ imds_ip4s() {
if [ "$IFACE" != eth0 ] && [ -n "$ip4s" ] &&
[ -z $(ip +F -4 route show table "$RTABLE" 2>/dev/null) ]; then
ip4p=$(echo "$ip4" | head -1) # primary IPv4
ip4_cidr=$(imds "@nic:$IFACE,@ipv4-net") # TODO: get from iface instead?
ip4_cidr=$(imds -E "@nic:$IFACE,@ipv4-net") # TODO: get from iface instead?
# TODO: this may not hold true for non-AWS clouds
ip4_gw=$(echo "$ip4_cidr" | cut -d/ -f1 |
awk -F. '{ print $1"."$2"."$3"."$4+1 }')
@ -73,7 +73,7 @@ imds_ip4s() {
# circle back and see how amazon-ec2-net-utils is handling everything these days
imds_ip6s() {
local ip6s gw tries=20
ip6s=$(imds "@nic:$IFACE,@ipv6")
ip6s=$(imds -E "@nic:$IFACE,@ipv6")
# non-eth0 interfaces need custom route tables
#

View File

@ -46,13 +46,17 @@ while true; do
esac
exit 0;;
-s|--setup) # just openrc for now
for phase in -boot -early -main -final '' -net; do
rc-update -a del "tiny-cloud$phase" || true
# for installing in mounted volumes
: "${ROOT:=}"
# start with a clean slate
log -s info " - tiny-cloud* services removed from all runlevels"
rm -f "$ROOT"/etc/runlevels/*/tiny-cloud*
log -s info " + tiny-cloud-boot service added to boot runlevel"
ln -s /etc/init.d/tiny-cloud-boot "$ROOT"/etc/runlevel/boot
for p in early main final; do
log -s info " + tiny-cloud-$p service added to default runlevel"
ln -s "/etc/init.d/tiny-cloud-$p" "$ROOT"/etc/runlevel/default
done
rc-update add tiny-cloud-boot boot
rc-update add tiny-cloud-early default
rc-update add tiny-cloud-main default
rc-update add tiny-cloud-final default
exit 0;;
--) shift; break;;
*) usage >&2; exit 1;;