Root layout, monitoring alerts, freeze hardening - #34
Merged
Merged
Conversation
chains/default/ now owns the shared runtime configuration: genesis-template.json, subnet-config.json, node-config.json, and the three chain-config variants. Resolution for every shipped file is the chain's own chains/<name>/ file, then chains/default/, then the legacy root name, so deployment roots cut before this layout keep working. default joins oracle and management as a reserved chain name. CONSENSUS-TUNING.md moves to docs/. nodes.ini.example joins the other inventory shapes in examples/. pack and bundle ship chains/, examples/, and docs/ instead of a root file list.
Prometheus evaluates monitoring/alerts.yml (fleet rules) plus one alerts.yml per installed app, mounted by the compose file. The rules are the trigger source for operations automation; the kit sends no notifications. No height-stall rule: the EVM produces blocks only when transactions arrive, so NodeBehindPeers covers the harmful case. A stale .gitignore rule from the generated-monitoring era kept monitoring/prometheus.yml out of git while the compose file, the bundle script, and playbook 06 all reference it. A fresh clone got a broken stack: Docker turns the missing file mount into a directory. The file is shipped now and the generated targets.json is ignored instead. All 11 rules pass promtool and evaluate healthy on the live fleet.
Monitoring: Prometheus alert rules, prometheus.yml enters git
owenwahlgren
changed the title
The root module was still named avalanche-benchmark/remote from the era when the Go code lived under remote/. The pack artifact carried the same relic (remote-benchmark.tar.gz). Mechanical rename, no logic change: module github.com/ava-labs/avalanche-benchmark, artifact avalanche-benchmark.tar.gz. This is the last chance before the production fork inherits the old path. go build ./..., go test ./..., gofmt all clean.
The pack artifact carried only the Grafana datasource and the dashboards. A packed control kit could not run playbook 06: no prometheus.yml, no alert rules, no compose file, no weight exporter, no provisioning config. The kit now carries the complete monitoring directory contents plus playbooks/ and docs/, so the control machine is self-documenting. Still no sources, still no secrets.
The repo had no Go CI; only Socket and CodeQL ran. The frozen POC needs a tripwire: any future PR gets gofmt, go build, and go test on ubuntu-latest. This also gives CodeQL's actions analyzer a workflow file to analyze, which removes one of its two no-source-found failures.
The manual began mid-thought. It now opens with the Avalanche logo (docs/avalanche-logo.png, from ava-labs/avalanchego resources), a one-line pitch, and a start-here table that routes each task to its playbook. The manual body is unchanged.
owenwahlgren
changed the title
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
The numbers encoded writing order, not reading order: rootless install and monitoring accreted at 05 and 06 while sitting mid-journey. New order: provision, rootless install, monitoring, load test, failover drill, validator swap, connected P-chain, install app, multi-chain. Monitoring moves before the drills because the dashboards should be up before anything breaks. Cross-references sweep: README table, playbook cross-links, alert-rule annotations.
PCHAIN_API_TOKEN rode every request to the public API host as a query argument matched by an infra-side rule. A public kit must not ship secret-token plumbing: an isolated deployment uses its own P-chain endpoint, and the public API works without a token at kit rates. Removed: the transport shim (internal/config/bypass.go), the .env field, the docs. The strict .env parser now rejects the field by name, so an old root gets a loud, clear error. tools/bundle.sh keeps blanking the field, because old deployment roots still carry it.
owenwahlgren
force-pushed
the
refactor/root-layout
branch
from
August 7, 2026 02:29
56f9dce to
773d4dd
Compare
MULTICHAIN-PLAN.md, PROPOSAL.md, and PACKAGE-SPEC.md live in the working tree by convention and never belong in a commit. Convention failed once; the ignore file makes it structural.
make pack writes avalanche-benchmark.tar.gz now; the ignore rule still named only the old artifact. Both names stay ignored: old roots carry the old tarball.
fleet_actual_weight carries machine/identity/dc/l1 labels; the annotation asked for a node label the exporter never emits and rendered empty.
CodeQL flagged the unrestricted GITHUB_TOKEN. The jobs only read the repository; contents: read is the whole requirement.
Review finding: subnetConfigPath and chainTemplatePath consulted chains/default/ before the oracle-specific legacy root names. An old oracle deployment root without chains/oracle/ would have silently picked up the generic consensus parameters instead of the oracle's own. Specificity now orders every layer: the chain's own file, the chain-specific legacy name, the shared default, the shared legacy name. Tests pin the ordering in both packages.
containerman17
approved these changes
Aug 7, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The repository root held six runtime config files, the consensus doc, and the inventory example. They now live where their concern lives. #35 (alert rules) merged into this branch, followed by four freeze-hardening commits. This PR is the last substantive change before the repo freezes as the POC.
What changed
Root layout:
chains/default/owns the shared runtime configuration:genesis-template.json,subnet-config.json,node-config.json, and the threechain-configvariants. Resolution for every shipped file is the chain's ownchains/<name>/file, thenchains/default/, then the legacy root name. Deployment roots cut before this layout keep working unchanged.defaultjoinsoracleandmanagementas a reserved chain name.CONSENSUS-TUNING.mdmoves todocs/.nodes.ini.examplejoins the other inventory shapes inexamples/.make packandtools/bundle.shshipchains/,examples/, anddocs/instead of a root file list; bundle still carries legacy root files when re-cutting an old deployment root.Monitoring alerts (#35):
monitoring/alerts.yml: 9 fleet alert rules.apps/settlement-feed/alerts.yml: 2 app rules. Apps ship alert rules next to their dashboards; the compose file mounts one file per app. No height-stall rule: the EVM produces blocks only on transactions, soNodeBehindPeerscovers the harmful case.monitoring/prometheus.ymlenters git. A stale.gitignorerule kept it out while the compose file, the bundle script, and playbook 06 all reference it; a fresh clone got a broken monitoring stack. The generatedtargets.jsonis ignored instead.Freeze hardening:
remote/relic (avalanche-benchmark/remote->avalanche-benchmark, 49 files, mechanical). The pack artifact follows:remote-benchmark.tar.gz->avalanche-benchmark.tar.gz. Last chance before the production fork inherits the old path.playbooks/anddocs/. A packed control kit could not run playbook 06 before this..github/workflows/go.ymlruns gofmt, build, and test on every PR. Also gives CodeQL's actions analyzer a workflow to analyze.Verification
go build ./...,go test ./...(15 packages), andgofmt -lclean after every commit, including the module rename.promtool check rulespasses (11 rules); deployed to the live Prometheus on the control host, all rules evaluatehealth=ok.Open questions
None.