CID in Deployment
Continuous Deployment was the second initial reason for FutoIn CID development.
CID is designed for zero-downtime rolling deployment with automatic migration support.
CID can be used standalone, but integration with provisioning systems is supported for maximum efficiency. A good reference integration is CodingFuture cfweb module.
Features
- Rolling deployment with zero downtime
- Automatic migration hooks (e.g. database schema update through flyway, liquibase, etc.)
- Resource limit auto-detection & distribution
- Multiple entry points per project
- File security enforcements (read-only)
Configuration hierarchy
The details of configuration tree and file locations are available in dedicated section.
For general understanding, there is the following hierarchy.
- Reasonable defaults of each tool.
- Global configuration in
/etc/futoin/futoin.json
. - User configuration in
${HOME}/.futoin.json
. - Project configuration in
${PROJECT_ROOT}/futoin.json
. - Deployment configuration in
${DEPLOY_ROOT}/futoin.json
.
Configuration is merged quite specific way for each entry. So, it’s not possible to describe it just as “deep merge”.
The most important part is that deployment configuration can override project defaults.
Deployment folder
A typical deployment folder:
current
- symlink to active deployed version{ver}
.persistent/
- folder or symlink to location for persistent file storage.vcs/
- VCS cache to optimize network usage of source deployments.-
{ver}
- VCS reference (tag or branch+revision) name or RMS package without extension.- active deployment
-
{ver}.tmp
- project version during deployment.- next active deployment after successful completion.
-
{ver}.{ext}
- package file for RMS deployments.- automatic cleanup process leaves only current active
{ver}
package for efficient forced re-deployment on demand.
- automatic cleanup process leaves only current active
-
.env
- optional file with environment configuration.- gets symlinked into
{ver}/.env
, if exists.
- gets symlinked into
.futoin-deploy.lock
- mandatory file as safety feature.futoin.json
- deployment configuration..futoin.merged.json
- merged project, deployment and environment configuration files with all default values set. Useful for provisioning systems..{any}
- any hidden ignored file or folder.
All other files are forcibly removed! Thefore, CID requires deployment target folder
to be completely empty or hold mandatory .futoin-deploy.lock
file as safety feature
Resource limits auto-detection
All resource limits are container-friendly (e.g. Docker) and automatically detected based on the following:
-
RAM:
--limit-memory
option is used, if present.- cgroup memory limit is used, if less than amount of RAM.
- half of RAM is used otherwise.
- Memory units: one of B, K, M, G postfixes is required. Example: 1G, 1024M, 1048576K, 1073741824B
-
CPU count:
--limit-cpus
option is used, if present.- cgroup CPU count is used, if present.
- all detected CPU cores are used otherwise.
-
Max clients:
- Auto-detected based on available memory and entry point configuration of
.connMemory
. - Can be used by load balancers and reverse-proxy servers.
- Auto-detected based on available memory and entry point configuration of
- File descriptor limit - auto-detected based on max clients and configured file descriptor count per client.
-
Instance count per entry point:
- if not
scalable
then only single instance is configured. - if not
multiCore
then: - get theoretical maximum of instances based on doubled
.minMemory
- get CPU limit count
- use
maxInstances
configuration, if any. - use the least value of detected above.
- otherwise, configure one instance.
- if not
Resource distribution & Entry Point instance auto-configuration
Based on overall resource limits per deployment, the resources are automatically distributed across entry points which may fine tune min/max memory, instance count, memory/cpu weights, per connection requirements and other constraints which affect auto-configuration.
As each instance should have some way to accept incoming connections, deployment process auto-assigned UNIX, TCP, UDP ports based on declared supported types.
Many tool support quite specific protocol like wsgi
or fcgi
- it affects auto-configuration
of web server, if any.
Entry points are described in more detail in dedicated section.
Zero-downtime deployment approach
This approach is used for classical, container and development deployments. However, actual zero-downtime benefit is assumed for “classical” non-container production case.
Step-by-step:
- a clean target folder is required for safety reasons due to automatic cleanup,
- deploy lock is taken on target folder,
-
target package:
- if
devserve
is used, the actual working copy is symlinked - if
vcsref
orvcsref
then local VCS cache is maintained for bandwidth efficiency - otherwise, last used RMS package is cached
- if
-
target version auto-detection:
- if
vcsref
is used then the latest revision is always used. - if precise version is set - it is used for deployment
- if partial package mask is set - it is used with shell-like match filtering
- for
rms
a list of available packages is retrieved efficient way - for
vcstag
a list of available tags is retrieved efficient way - the retrieved list of candidates is sorted in natural order (decimal numbers are assumed)
- the latest one (greatest by order) is used
- if
-
persistent data:
persistent
configuration is used to setup read-write persistent paths.- read-write location root is set to
{deployment root}/persistent/
by default. - if specified file or directory exists in package, it is forcibly copied to read-write location (!).
- otherwise, a folder is created in read-write location with symlink from target folder.
- it’s expected that persistent folder is subject for backup procedures.
- a temporary folder under deployment root is used,
-
the actions are executed:
- actions can be hooked both in project and deployment configuration:
.actions
is a map of named actions to string or list of commands.- Standard actions match some of command names: “prepare”, “build”, “migrate”, etc.
@cid
in the beginning of command is treated as CID invocation. Example:@cid build-dep openssl
@default
as command executes the default behavior. For deployment config it executes project-specified action configuration.- If command matches any of other defined actions then it is executed with recursion of this logic.
- Note: there is recursion protection other than program stack size.
- See
cid deploy set action
for easy scripting instead of direct JSON manipulations. - if VCS deployment or forced with
--build
option cid prepare
- suitable for extra setupcid build
cid migrate
- suitable for auto-configuration & database migrations
- all files and directories are set read-only for security & data safety purposes (enforce persistent locations),
- temporary folder is renamed to package name without extension, VCS tag or VCS branch with revision name,
current
symlink is set to above,-
if running
cid service master
is detected then it is refreshed,- note: very slight delay may occur which expected to be smoothed by load balancer?
-
deployment folder is cleaned out of any not expected files and folder (cleanup of old versions & misc.),
- note: there are some extra files & folders like .tmp, .runtime, .futoin-deploy.lock, etc.,
- deploy lock is released,
- at any point, if something goes wrong the procedure is aborted leaving previous version running as is.
Containers
CID can be easily used inside Docker or other type of image as it is cgroup-friendly. Service master mode and port allocation hints completely satisfy such use case.
systemd
Services and Services are auto-configured based on entry point configurations. Services can be either
executed by Service Master (cid service master
) or through individual service instance
execution (cid service exec
). Both cases work well with systemd
.
Commands
Migration
This action is implicitely executed during deployment. It is provided here for reference to be overridden in configuration.
-
cid migrate
- Runs data migration tasks.
- Provided for overriding default procedures in scope of deployment procedure.
Deployment operation
This is common set of commands used for deployment of project from source or build artifacts.
-
cid deploy ...
- Common arguments for deploy family of commands:
[--deployDir=<deploy_dir>]
- target folder, CWD by default.[--runtimeDir=<runtime_dir>]
- target runtime data folder,<deploy_dir>/.runtime
by default.[--tmpDir=<tmp_dir>]
- target temporary data folder,<deploy_dir>/.tmp
by default.[--limit-memory=<mem_limit>]
- memory limit with B, K, M or G postfix.[--limit-cpus=<cpu_count>]
- max number of CPU cores to use.[--listen-addr=<address>]
- address to use for IP services[--user=<user>]
- user name to run services.[--group=<group>]
- group name to run services.
-
cid deploy setup
- Prepare directory for deployment. Allows adjusting
futoin.json
before actual deployment is done to define limits once or add project settings overrides. Allows adjusting settings for next deployment. Not necessary otherwise.
- Prepare directory for deployment. Allows adjusting
-
cid deploy vcstag [<vcs_ref>] [--vcsRepo=<vcs_repo>] [--redeploy]
- Deploy from VCS tag.
-
cid deploy vcsref <vcs_ref> [--vcsRepo=<vcs_repo>] [--redeploy]
- Deploy from VCS branch.
-
cid deploy rms <rms_pool> [<package>] [--rmsRepo=<rms_repo>] [--build]
- Deploy from RMS.
Deployment configuration
Configuration overriding in deployment target folder is supported. Primary case
is for CID-unaware projects. It’s possible to do with direct manipulation of futoin.json
.
However, the commands below much better suit for scripts then direct manipulation. Please check Examples section.
-
cid deploy set ...
-
Common arguments for “deploy set” family of commands:
- [—deployDir=<deploy_dir>] - target folder, CWD by default.
-
-
cid deploy set tools <tools>...
- Overrides .tools in deployment config.
-
cid deploy set tooltune <tool> {<set_name=value>...|<del_name>|<inline_json>}
- Override .toolTune in deployment config.
-
cid deploy set action <name> <actions>...
- Override .action in deployment config.
-
cid deploy set persistent <paths>...
- Add .persistent paths in deployment config.
-
cid deploy set entrypoint <name> <tool> <path> {<set_name=value>...|<del_name>|<inline_json>}
- Set entry point configuration in deployment.
-
cid deploy set env <variable> [<value>]
- Set or remote environment config .env entries.
cid deploy set webcfg <variable> [<value>]
-
cid deploy set webcfg mounts <route>[=<app>]
- Set or remove .webcfg entries.
-
cid deploy set webmount <web_path> [<json>]
- Set complex web mount point configuration.
-
cid deploy reset [<set_type>]
- Reset related configuration to initial state.
- Use any known “deploy set” type in place of
<set_type>
. - Useful for automation to ensure a clean state.
Service execution
This service execution commands are designed for production use.
Please use cid devserve
for development.
-
cid service ...
- Service execution helpers common arguments:
[--deployDir=<deploy_dir>]
- target folder, CWD by default.
-
cid service master [--adapt [*generic deploy options*]]
- Re-balance services, if
--adapt
. - Run all entry points as children.
- Restarts services on exit.
- Has 10 second delay for too fast to exit services.
- Supports SIGTERM for clean shutdown.
- Supports SIGHUP for reload of service list & the services themselves.
- Re-balance services, if
-
cid service list [--adapt [*generic deploy options*]]
- Re-balance services, if
--adapt
. - List services in the following format:
<entry point> <TAB> <instance ID> <TAB> <socket type> <TAB> <socket address>
- Re-balance services, if
-
cid service exec <entry_point> <instance_id>
- Helper for system init to execute pre-configured service.
-
cid service stop <entry_point> <instance_id> <pid>
- Helper for system init to gracefully stop pre-configured service.
-
cid service reload <entry_point> <instance_id> <pid>
- Helper for system init to gracefully reload pre-configured service.
- Note: if reload is not supported then reload acts as “stop” to force restart.
Execution in Development
The ultimate goal is to provide the same execution environment as in production. Otherwise, it’s quite often appears that projects work well with so-called “development servers”, but appear to be not ready for proper web servers like “nginx”.
At the moment, cid devserve
is not related to cid build
procedure. Therefore, the
commands must be called independently. Incremental builds may get native support as well.
-
cid devserve [--wcDir=<wc_dir>] [*generic deploy options*]
- Run as closely to production deployment as possible.
- Create temporary deployment directory and use working directory as “current”.
- Re-balance services.
- Then act like
cid service list
andcid service master
.