Repository navigation
Guided installer for server deployments (version v 0.2.0) - #165
paolocappelletti wants to merge 1 commit into
Conversation
saratnt
left a comment
saratnt
left a comment
There was a problem hiding this comment.
From Claude:
_pkg/manager/config.go:190 builds the URL as exactly:
protocol + "://" + address + ":" + port
In case the address contians a path, the port lands inside the path. Eg
https://horizen-testnet.rpc.caldera.xyz/http:443 instead of https://horizen-testnet.rpc.caldera.xyz:443/http
| | | `horizen-l3-testnet` | `base-sepolia` | | ||
| |---|---|---| | ||
| | chainId | 2651420 | 84532 | | ||
| | RPC | `https://horizen-testnet.rpc.caldera.xyz` | `https://sepolia.base.org` | |
There was a problem hiding this comment.
Shouldn't be https://horizen-testnet.rpc.caldera.xyz/http?
| NET_RPC_PROTOCOL="https" | ||
| NET_RPC_HOST="horizen-testnet.rpc.caldera.xyz" | ||
| NET_RPC_PORT="443" | ||
| NET_RPC_URL="https://horizen-testnet.rpc.caldera.xyz" |
There was a problem hiding this comment.
Shouldn't be https://horizen-testnet.rpc.caldera.xyz/http?
There was a problem hiding this comment.
privileged: true + devices: /dev/vsock survive into a TCP-only bundle.
Both executor and manager keep the vsock device mapping, but the bundle forces CHANNEL_TYPE='tcp' and no-TEE. On a server without /dev/vsock (the common case for a plain VM), docker compose up fails at container creation with error gathering device information while adding custom device "/dev/vsock". README-DEPLOY.md troubleshoots port and subnet collisions but not this. Strip devices (and privileged, which is unnecessary here) for CHANNEL_TYPE=tcp, or document modprobe vhost_vsock as a server prerequisite.
| warn "RPC $method returned an error: $resp" | ||
| return 1 | ||
| fi | ||
| printf '%s' "$resp" | sed -n 's/.*"result"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p' |
There was a problem hiding this comment.
rpc_result exits 0 with empty stdout when its sed doesn't match the response shape (eg in case of a HTML 404 error). Then some checks don't work as expected (eg at line 316, 652, 725)
| #----------------------------------------------------------------------------- | ||
| readonly SCRIPT_NAME="$(basename "$0")" | ||
| readonly REPO_SLUG="HorizenOfficial/vela" | ||
| readonly SUPPORTED_VERSIONS=("0.2.0") |
There was a problem hiding this comment.
We should remember at any new release to update this variable. Claude suggests one of the following options:
- Minimal — extend the CLAUDE.md note with the release-time trigger: "Cutting a release requires adding the new version to SUPPORTED_VERSIONS in install-helper/vela-install.sh; the installer offers a closed menu and cannot install a tag that is not listed."
- Enforced — add a step to the existing v* tag job asserting ${GITHUB_REF_NAME#v} appears in SUPPORTED_VERSIONS. Cheap, and it fails at exactly the right moment.
- Structural — derive the list instead of hardcoding it (GitHub's tag list, or the ## headings in CHANGELOG.md), and keep the shape assertions as the real safety net, since they already stop the run with an explicit error when an unfamiliar tag doesn't match.
| // volume mounts of dropped volumes (e.g. horizen-cce-deploy-data:/deploy-data). | ||
| bl = filterKeyBlock(bl, 4, 'volumes', (name) => !dropVolumes.has(name)); | ||
| // "profiles: [...]" only exists on kms-proxy, but drop defensively. | ||
| bl = dropKeyBlock(bl, 4, 'profiles'); |
There was a problem hiding this comment.
dropKeyBlock(bl, 4, 'profiles') matches only a bare profiles: line; upstream uses the inline form profiles: ["kms"]. Dead today (kms-proxy is dropped wholesale), but it won't protect a kept service that later gains a profile.
Guided installer script for server deployments
(Since it is based on 0.2.0 has been openend on master branch)