|
| 1 | +# Decision: Cloud-Init SSH Port Configuration with Reboot |
| 2 | + |
| 3 | +## Status |
| 4 | + |
| 5 | +Accepted |
| 6 | + |
| 7 | +## Date |
| 8 | + |
| 9 | +2025-12-11 |
| 10 | + |
| 11 | +## Context |
| 12 | + |
| 13 | +The deployer needs to support custom SSH ports for security and flexibility. The SSH port configuration must be applied **during VM provisioning** (not later in the configure phase) because: |
| 14 | + |
| 15 | +1. **Provision phase dependencies**: The `WaitForCloudInitStep` runs during provision and uses Ansible to wait for cloud-init completion. Ansible connects using the custom port from the inventory configuration. |
| 16 | + |
| 17 | +2. **Timing requirement**: If SSH is not already listening on the custom port when `WaitForCloudInitStep` executes, the provision command fails with connection errors. |
| 18 | + |
| 19 | +3. **Architectural correctness**: SSH port is infrastructure configuration, not application configuration. It should be set during infrastructure provisioning, not as a post-provisioning step. |
| 20 | + |
| 21 | +The challenge was ensuring SSH service reliably restarts with the new port configuration during cloud-init execution. Multiple approaches were tested: |
| 22 | + |
| 23 | +- **systemctl restart**: Does not kill the old SSH process when port changes, resulting in SSH listening on both ports 22 and the custom port |
| 24 | +- **pkill + systemctl start**: Works but is brittle and non-standard |
| 25 | +- **bootcmd disable + runcmd restart**: Ineffective because systemd automatically re-enables and starts SSH after bootcmd completes |
| 26 | + |
| 27 | +## Decision |
| 28 | + |
| 29 | +We configure the custom SSH port via cloud-init using the **`write_files` + `reboot` pattern**, following Hetzner's cloud-config best practices: |
| 30 | + |
| 31 | +1. **Write SSH configuration file** using cloud-init's `write_files` directive: |
| 32 | + |
| 33 | + ```yaml |
| 34 | + {% if ssh_port != 22 %} |
| 35 | + write_files: |
| 36 | + - path: /etc/ssh/sshd_config.d/99-custom-port.conf |
| 37 | + content: | |
| 38 | + # Custom SSH port configuration |
| 39 | + Port {{ ssh_port }} |
| 40 | + permissions: "0644" |
| 41 | + owner: root:root |
| 42 | + ``` |
| 43 | +
|
| 44 | +2. **Trigger system reboot** in cloud-init's `runcmd` phase: |
| 45 | + |
| 46 | + ```yaml |
| 47 | + runcmd: |
| 48 | + - reboot |
| 49 | + {% endif %} |
| 50 | + ``` |
| 51 | + |
| 52 | +**Conditional Configuration**: The SSH port configuration and reboot only execute when `ssh_port != 22`, avoiding unnecessary reboots for environments using the default SSH port. |
| 53 | + |
| 54 | +The reboot ensures: |
| 55 | + |
| 56 | +- SSH service cleanly restarts with the new configuration |
| 57 | +- No old SSH processes remain on port 22 |
| 58 | +- All services start in a consistent state |
| 59 | +- Package updates are applied (if cloud-init installed packages) |
| 60 | + |
| 61 | +Additionally, we made two critical fixes to the provision handler: |
| 62 | + |
| 63 | +1. **Use configured SSH port**: Changed `wait_for_readiness()` to use `SocketAddr::new(ip, ssh_port)` instead of `SshConfig::with_default_port()`, ensuring the provision handler waits for SSH on the correct custom port (not port 22). |
| 64 | + |
| 65 | +2. **Increase SSH connectivity timeout**: Raised `DEFAULT_MAX_RETRY_ATTEMPTS` from 30 to 60 attempts (120 seconds total), accounting for the ~70-80 second cloud-init completion time plus reboot time. |
| 66 | + |
| 67 | +## Consequences |
| 68 | + |
| 69 | +### Positive |
| 70 | + |
| 71 | +- **Clean SSH restart**: Reboot guarantees SSH only listens on the custom port, no lingering processes on port 22 |
| 72 | +- **Industry best practice**: Follows Hetzner's documented cloud-config pattern for SSH port changes |
| 73 | +- **Simple and reliable**: Single `reboot` command is simpler than managing service lifecycle manually |
| 74 | +- **Correct architecture**: Infrastructure configuration happens during infrastructure provisioning |
| 75 | +- **No special cases**: Ansible can connect normally using the configured port without overrides or workarounds |
| 76 | +- **Compile-time safety**: Provision handler correctly waits for the configured port, preventing connection failures |
| 77 | +- **Conditional execution**: Only reboots when custom port is needed (ssh_port != 22), avoiding unnecessary reboots for default configurations |
| 78 | + |
| 79 | +### Negative |
| 80 | + |
| 81 | +- **Slower provisioning**: Reboot adds ~10-20 seconds to VM initialization time |
| 82 | +- **Additional wait time**: Provision handler must wait longer (120s instead of 60s) for cloud-init and reboot to complete |
| 83 | +- **Complexity**: Three separate changes required (cloud-init template, provision handler port usage, timeout increase) |
| 84 | + |
| 85 | +### Risks |
| 86 | + |
| 87 | +- **Reboot timing**: If reboot takes longer than expected, SSH connectivity check might timeout (mitigated by 120-second timeout) |
| 88 | +- **Cloud-init failure**: If reboot fails or cloud-init has errors, the provision will fail (acceptable - we want to catch infrastructure issues early) |
| 89 | + |
| 90 | +## Alternatives Considered |
| 91 | + |
| 92 | +### Alternative 1: Ansible Playbook in Configure Phase |
| 93 | + |
| 94 | +**Approach**: Use an Ansible playbook during the `configure` phase to reconfigure SSH port after provisioning. |
| 95 | + |
| 96 | +**Why Rejected**: |
| 97 | + |
| 98 | +- **Timing problem**: `WaitForCloudInitStep` in provision already fails before reaching configure phase |
| 99 | +- **Architectural mismatch**: SSH port is infrastructure config, should be set during VM initialization |
| 100 | +- **Added complexity**: Requires special connection handling (connect on 22, reconfigure, reconnect on custom port) |
| 101 | +- **More failure points**: Port transition adds potential for connection issues |
| 102 | + |
| 103 | +### Alternative 2: systemctl restart Without Reboot |
| 104 | + |
| 105 | +**Approach**: Use cloud-init `runcmd` to execute `systemctl restart ssh` without full system reboot. |
| 106 | + |
| 107 | +**Why Rejected**: |
| 108 | + |
| 109 | +- **Doesn't kill old process**: `systemctl restart` doesn't terminate the existing SSH daemon when port changes |
| 110 | +- **Dual port listening**: Results in SSH listening on both port 22 (old) and custom port (new) |
| 111 | +- **Testing showed failure**: Multiple test attempts confirmed SSH remained on port 22 after cloud-init "completion" |
| 112 | + |
| 113 | +### Alternative 3: pkill + systemctl start |
| 114 | + |
| 115 | +**Approach**: Kill SSH processes with `pkill -9 sshd`, then start fresh with `systemctl start ssh`. |
| 116 | + |
| 117 | +**Why Rejected**: |
| 118 | + |
| 119 | +- **Non-standard**: Violates best practices for service management |
| 120 | +- **Brittle**: Process killing is less reliable than clean reboot |
| 121 | +- **Not industry pattern**: No documentation or precedent for this approach |
| 122 | + |
| 123 | +### Alternative 4: Wait for Port 22, Then Handle Port Change |
| 124 | + |
| 125 | +**Approach**: Keep provision handler waiting for port 22, handle port transition separately. |
| 126 | + |
| 127 | +**Why Rejected**: |
| 128 | + |
| 129 | +- **Wrong abstraction**: Provision handler should use the configured port, not hardcode defaults |
| 130 | +- **Added complexity**: Would require special logic to detect port changes mid-provision |
| 131 | +- **Race conditions**: SSH might move to custom port at unpredictable times during cloud-init |
| 132 | + |
| 133 | +## Related Decisions |
| 134 | + |
| 135 | +- [Register Command SSH Port Override](./register-ssh-port-override.md) - Relates to SSH port handling in different commands |
| 136 | +- [Environment Variable Prefix](./environment-variable-prefix.md) - Relates to configuration management patterns |
| 137 | + |
| 138 | +## References |
| 139 | + |
| 140 | +- [Hetzner Cloud-Config Tutorial](https://community.hetzner.com/tutorials/basic-cloud-config) - Section 5.3 documents the reboot pattern for SSH configuration |
| 141 | +- [Cloud-Init Documentation](https://cloudinit.readthedocs.io/en/latest/) - Official cloud-init reference |
| 142 | +- [Issue #222: Configure SSH Service Port](../issues/222-configure-ssh-service-port.md) - Original issue specification |
| 143 | +- [OpenSSH sshd_config.d](https://manpages.debian.org/bookworm/openssh-server/sshd_config.5.en.html#Include) - Ubuntu SSH configuration directory pattern |
0 commit comments