diff options
Diffstat (limited to 'users')
-rw-r--r-- | users/edef/weave/Cargo.nix | 2 | ||||
-rw-r--r-- | users/flokli/nixos/.envrc | 2 | ||||
-rw-r--r-- | users/flokli/nixos/default.nix | 8 | ||||
-rw-r--r-- | users/flokli/nixos/nixos-tvix-cache/OWNERS | 1 | ||||
-rw-r--r-- | users/flokli/nixos/nixos-tvix-cache/README.md | 40 | ||||
-rw-r--r-- | users/flokli/nixos/nixos-tvix-cache/configuration.nix | 81 | ||||
-rw-r--r-- | users/flokli/nixos/nixos-tvix-cache/disko.nix | 62 | ||||
-rw-r--r-- | users/flokli/nixos/nixos-tvix-cache/monitoring.nix | 163 | ||||
-rw-r--r-- | users/flokli/nixos/nixos-tvix-cache/nar-bridge-module.nix | 75 | ||||
-rw-r--r-- | users/flokli/nixos/nixos-tvix-cache/nar-bridge.nix | 141 |
10 files changed, 574 insertions, 1 deletions
diff --git a/users/edef/weave/Cargo.nix b/users/edef/weave/Cargo.nix index ea1d8a1b2ef3..c4a52ad1b7cd 100644 --- a/users/edef/weave/Cargo.nix +++ b/users/edef/weave/Cargo.nix @@ -7193,7 +7193,7 @@ rec { ]; features = { "axum" = [ "dep:axum" ]; - "otlp" = [ "dep:tracing-opentelemetry" "dep:opentelemetry" "dep:opentelemetry-otlp" "dep:opentelemetry_sdk" "dep:opentelemetry-http" "reqwest-tracing?/opentelemetry_0_22" ]; + "otlp" = [ "dep:tracing-opentelemetry" "dep:opentelemetry" "dep:opentelemetry-otlp" "dep:opentelemetry_sdk" "dep:opentelemetry-http" "dep:opentelemetry-semantic-conventions" "reqwest-tracing?/opentelemetry_0_27" ]; "reqwest" = [ "dep:reqwest-tracing" ]; "tonic" = [ "dep:tonic" "dep:http" ]; "tracy" = [ "dep:tracing-tracy" ]; diff --git a/users/flokli/nixos/.envrc b/users/flokli/nixos/.envrc index ccf3cb847ac5..5c14eb6e9edd 100644 --- a/users/flokli/nixos/.envrc +++ b/users/flokli/nixos/.envrc @@ -1 +1,3 @@ +source_up + PATH_add $(nix-build ../../.. -A users.flokli.nixos.deps --no-out-link)/bin diff --git a/users/flokli/nixos/default.nix b/users/flokli/nixos/default.nix index cce4801f45bd..b507659d94b4 100644 --- a/users/flokli/nixos/default.nix +++ b/users/flokli/nixos/default.nix @@ -21,6 +21,14 @@ depot.nix.readTree.drvTargets rec { deploy-archeology-ec2 = (deployScript "archeology-ec2" archeologyEc2System); + nixosTvixCacheSystem = (depot.ops.nixos.nixosFor ({ ... }: { + imports = [ + ./nixos-tvix-cache/configuration.nix + ]; + })).config.system.build.toplevel; + + deploy-nixos-tvix-cache = (deployScript "root@nixos.tvix.store" nixosTvixCacheSystem); + deps = (depot.nix.lazy-deps { deploy-archeology-ec2.attr = "users.flokli.nixos.deploy-archeology-ec2"; aws.attr = "third_party.nixpkgs.awscli"; diff --git a/users/flokli/nixos/nixos-tvix-cache/OWNERS b/users/flokli/nixos/nixos-tvix-cache/OWNERS new file mode 100644 index 000000000000..d96354420040 --- /dev/null +++ b/users/flokli/nixos/nixos-tvix-cache/OWNERS @@ -0,0 +1 @@ +zimbatm diff --git a/users/flokli/nixos/nixos-tvix-cache/README.md b/users/flokli/nixos/nixos-tvix-cache/README.md new file mode 100644 index 000000000000..6a0e2f8ddbd8 --- /dev/null +++ b/users/flokli/nixos/nixos-tvix-cache/README.md @@ -0,0 +1,40 @@ +# nixos-tvix-cache + +This is a fetch-through mirror of cache.nixos.org, hosted by NumTide. + +The current machine is a SX65 Hetzner dedicated server with 4x22TB SATA disks, +and 2x1TB NVMe disks. + +The goals of this machine: + + - Exercise tvix-store and nar-bridge code + - Collect usage metrics (see [Grafana](https://nixos.tvix.store/grafana)) + - Identify bottlenecks in the current implementations and fix them + - Replace cache.nixos.org? + +You can configure this as a Nix substitutor on your systems like this: + +```nix + nix.settings.substituters = lib.mkForce [ + "https://nixos.tvix.store" + ]; +``` + +For store paths it hasn't already seen yet, it'll internally ingest its contents +into tvix-castore (deduplicating in doing so). + +Requests for NARs will dynamically reassemble the NAR representation on demand. + +Metadata and signatures are preserved (which is why you don't need to add +additional trusted keys). +We need to produce the same data bit by bit, else the signature check in your +Nix/Lix client would fail. + +Be however aware that there's zero availability guarantees. +We will frequently redeploy this box, and it might become unavailable without +prior notice. + +Tvix currently doesn't have garbage collection. If we run out of disk space, we +might either move things to a bigger box or delete everything on it so far. + +As it's only a cache, it should however re-ingest things again. diff --git a/users/flokli/nixos/nixos-tvix-cache/configuration.nix b/users/flokli/nixos/nixos-tvix-cache/configuration.nix new file mode 100644 index 000000000000..221ea7f8779f --- /dev/null +++ b/users/flokli/nixos/nixos-tvix-cache/configuration.nix @@ -0,0 +1,81 @@ +{ pkgs +, lib +, config +, ... +}: + +let + srvos = + import (builtins.fetchTarball { + url = "https://github.com/nix-community/srvos/archive/15b152766b329dd2957549a49f0fd96a7a861db1.tar.gz"; + sha256 = "sha256-11TCdlxJEf84Lm2KIJGL8J2nJ2G9CNTW8PrCebJLg/M="; + }); + disko = + (builtins.fetchTarball { + url = "https://github.com/nix-community/disko/archive/84dd8eea9a06006d42b8af7cfd4fda4cf334db81.tar.gz"; + sha256 = "13mfnjnjp21wms4mw35ar019775qgy3fnjc59zrpnqbkfmzyvv02"; + }); + + +in +{ + imports = [ + "${disko}/module.nix" + ./disko.nix + ./monitoring.nix + ./nar-bridge.nix + srvos.nixosModules.hardware-hetzner-online-amd + srvos.nixosModules.mixins-nginx + ]; + + options = { + machine.domain = lib.mkOption { + type = lib.types.str; + default = "nixos.tvix.store"; + }; + }; + + config = { + services.nginx.virtualHosts."${config.machine.domain}" = { + enableACME = true; + forceSSL = true; + }; + + + security.acme.acceptTerms = true; + security.acme.defaults.email = "admin+acme@numtide.com"; + + nixpkgs.hostPlatform = "x86_64-linux"; + + networking.hostName = "tvix-cache"; + + systemd.network.networks."10-uplink".networkConfig.Address = "2a01:4f9:3071:1091::2/64"; + + + # Enable SSH and add some keys + services.openssh.enable = true; + users.users.root.openssh.authorizedKeys.keys = [ + # edef + "cert-authority ssh-rsa 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 edef" + # flokli + "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIPTVTXOutUZZjXLB0lUSgeKcSY/8mxKkC0ingGK1whD2 flokli" + # mic92 + "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIKbBp2dH2X3dcU1zh+xW3ZsdYROKpJd3n13ssOP092qE" + "ecdsa-sha2-nistp256 AAAAE2VjZHNhLXNoYTItbmlzdHAyNTYAAAAIbmlzdHAyNTYAAABBBCsjXKHCkpQT4LhWIdT0vDM/E/3tw/4KHTQcdJhyqPSH0FnwC8mfP2N9oHYFa2isw538kArd5ZMo5DD1ujL5dLk= ssh@secretive.Joerg’s-Laptop.local" + # padraic + "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIEFlro/QUDlDpaA1AQxdWIqBg9HSFJf9Cb7CPdsh0JN7" + # zimbatm + "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIOuiDoBOxgyer8vGcfAIbE6TC4n4jo8lhG9l01iJ0bZz zimbatm@no1" + "sk-ssh-ed25519@openssh.com AAAAGnNrLXNzaC1lZDI1NTE5QG9wZW5zc2guY29tAAAAINwWC6CJ/E6o3WGeZxbZMajC4roXnzVi8fOo1JYJSE6YAAAABHNzaDo= zimbatm@nixos" + ]; + + environment.systemPackages = [ + pkgs.helix + pkgs.htop + pkgs.kitty.terminfo + pkgs.tmux + ]; + + system.stateVersion = "24.11"; + }; +} diff --git a/users/flokli/nixos/nixos-tvix-cache/disko.nix b/users/flokli/nixos/nixos-tvix-cache/disko.nix new file mode 100644 index 000000000000..ef7c8def3daa --- /dev/null +++ b/users/flokli/nixos/nixos-tvix-cache/disko.nix @@ -0,0 +1,62 @@ +# Disk /dev/nvme0n1: 1024 GB (=> 953 GiB) +# Disk /dev/nvme1n1: 1024 GB (=> 953 GiB) +# Disk /dev/sda: 22 TB (=> 20 TiB) +# Disk /dev/sdb: 22 TB (=> 20 TiB) +# Disk /dev/sdc: 22 TB (=> 20 TiB) +# Disk /dev/sdd: 22 TB (=> 20 TiB) +# +# # Installation +# 1. Comment out the fileSystems +# 2. Bootstrap the machine with `clan machines tvix-cache-install` +# 3. Do the btrfs partitioning by hand (because it's not supported by Disko) +# a. `mkfs.btrfs -m raid1 -d single /dev/sd{a,b,c,d} --label tank -f` +# b. `mkdir /tank && mount /dev/disk/by-label/tank /tank` +# 4. Uncomment the fileSystems section below +# 5. Re-deploy +# +# TODO: make use of /dev/nvme1n1 +{ + boot.loader.efi.canTouchEfiVariables = true; + boot.loader.systemd-boot.configurationLimit = 10; + boot.loader.systemd-boot.enable = true; + boot.loader.timeout = 3; + boot.supportedFilesystems = [ "btrfs" ]; + + # TODO: comment me during install + fileSystems."/tank" = { + fsType = "btrfs"; + device = "/dev/disk/by-label/tank"; + }; + + disko.devices = { + disk = { + main = { + type = "disk"; + device = "/dev/nvme0n1"; + content = { + type = "gpt"; + partitions = { + ESP = { + size = "1G"; + type = "EF00"; + content = { + type = "filesystem"; + format = "vfat"; + mountpoint = "/boot"; + mountOptions = [ "umask=0077" ]; + }; + }; + root = { + size = "100%"; + content = { + type = "filesystem"; + format = "btrfs"; + mountpoint = "/"; + }; + }; + }; + }; + }; + }; + }; +} diff --git a/users/flokli/nixos/nixos-tvix-cache/monitoring.nix b/users/flokli/nixos/nixos-tvix-cache/monitoring.nix new file mode 100644 index 000000000000..d7c54c6a1434 --- /dev/null +++ b/users/flokli/nixos/nixos-tvix-cache/monitoring.nix @@ -0,0 +1,163 @@ +{ config, pkgs, ... }: +let + domain = config.machine.domain; +in +{ + # Configure the NixOS machine with Grafana and Tempo to collect metrics from nar-bridge. + + services.tempo = { + enable = true; + settings = { + auth_enabled = false; + server = { + http_listen_address = "127.0.0.1"; + http_listen_port = 9080; + grpc_listen_address = "127.0.0.1"; + grpc_listen_port = 9095; + grpc_server_max_recv_msg_size = 67108864; + grpc_server_max_send_msg_size = 67108864; + log_level = "warn"; + }; + + # move the otlp listener to another port than 4317, and disable the 4318 one. + # opentelemetry-connector binds on both 4317 and 4318. + distributor.receivers.otlp.protocols = { + grpc.endpoint = "127.0.0.1:4319"; + }; + + storage.trace = { + backend = "local"; + wal.path = "/var/lib/tempo/wal"; + local.path = "/var/lib/tempo/blocks"; + }; + usage_report.reporting_enabled = false; + # 10x the default + overrides.defaults.ingestion.max_traces_per_user = 10000 * 10; + }; + }; + + services.alloy.enable = true; + + environment.etc."alloy/config.alloy".text = '' + // Accept OTLP. Forward metrics to victoriametrics, and traces to tempo. + otelcol.receiver.otlp "main" { + grpc { + endpoint = "[::1]:4317" + } + + http { + endpoint = "[::1]:4318" + } + + output { + metrics = [otelcol.exporter.otlphttp.victoriametrics.input] + traces = [otelcol.exporter.otlp.tempo.input] + } + } + + // We push to Tempo over otlp-grpc. + otelcol.exporter.otlp "tempo" { + client { + endpoint = "127.0.0.1:4319" + tls { + insecure = true + } + } + } + + // We push to VictoriaMetrics over otlp-http. + otelcol.exporter.otlphttp "victoriametrics" { + client { + endpoint = "http://localhost:8428/opentelemetry" + } + } + + // Run a bundled node-exporter. + prometheus.exporter.unix "main" { } + + // Scrape it. + prometheus.scrape "main" { + targets = prometheus.exporter.unix.main.targets + forward_to = [otelcol.receiver.prometheus.default.receiver] + } + + // Convert Prometheus metrics to OTLP and export them. + otelcol.receiver.prometheus "default" { + output { + metrics = [otelcol.exporter.otlphttp.victoriametrics.input] + } + } + ''; + + services.victoriametrics.enable = true; + + + services.grafana = { + enable = true; + + settings = { + server = { + domain = domain; + http_addr = "127.0.0.1"; + http_port = 3000; + root_url = "https://%(domain)s/grafana"; + serve_from_sub_path = true; + }; + analytics.reporting_enabled = false; + "auth.anonymous" = { + enabled = true; + }; + auth.disable_login_form = true; + "auth.basic".enabled = false; + "auth.github" = { + enabled = true; + client_id = "Ov23liAnuBwzWtJJ7gv4"; + client_secret = "$__file{/run/credentials/grafana.service/github_auth_client_secret}"; + scopes = "user:email,read:org"; + auth_url = "https://github.com/login/oauth/authorize"; + token_url = "https://github.com/login/oauth/access_token"; + api_url = "https://api.github.com/user"; + allow_sign_up = true; + auto_login = false; + allowed_organizations = [ "numtide" ]; + role_attribute_path = "contains(groups[*], '@numtide/network') && 'GrafanaAdmin' || 'Viewer'"; + }; + }; + + provision = { + enable = true; + datasources.settings.datasources = [ + { + name = "Tempo"; + type = "tempo"; + uid = "traces"; + url = "http://127.0.0.1:9080"; + access = "proxy"; + timeout = "300"; + + jsonData = { + nodeGraph.enabled = true; + # tracesToLogs.datasourceUid = "logs"; + tracesToMetrics.datasourceUid = "metrics"; + # serviceMap.datasourceUid = "metrics"; + # nodeGraph.enabled = true; + # lokiSearch.datasourceUid = "logs"; + }; + } + { + name = "prometheus"; + type = "prometheus"; + uid = "metrics"; + url = "http://localhost:8428/"; + } + ]; + }; + }; + + systemd.services.grafana.serviceConfig.LoadCredential = "github_auth_client_secret:/etc/secrets/grafana_github_auth_client_secret"; + + services.nginx.virtualHosts."${domain}".locations."/grafana" = { + proxyPass = "http://localhost:3000"; + proxyWebsockets = true; + }; +} diff --git a/users/flokli/nixos/nixos-tvix-cache/nar-bridge-module.nix b/users/flokli/nixos/nixos-tvix-cache/nar-bridge-module.nix new file mode 100644 index 000000000000..3370f1aaa3cb --- /dev/null +++ b/users/flokli/nixos/nixos-tvix-cache/nar-bridge-module.nix @@ -0,0 +1,75 @@ +{ config +, lib +, pkgs +, depot +, ... +}: +let + cfg = config.services.nar-bridge; + + package = depot.tvix.nar-bridge.override (old: { + features = old.features or [ "default" ] ++ [ "xp-store-composition-cli" ]; + runTests = true; + }); + + storeCompositionFormat = pkgs.formats.toml { }; + + storeCompositionFile = storeCompositionFormat.generate "store-composition.toml" cfg.settings; + + args = [ + "--listen-address" + "sd-listen" + "--experimental-store-composition" + storeCompositionFile + ]; +in +{ + options = { + services.nar-bridge = { + enable = lib.mkEnableOption "nar-bridge service"; + + settings = lib.mkOption { + type = storeCompositionFormat.type; + default = { }; + }; + }; + }; + + config = lib.mkIf cfg.enable { + users.users.nar-bridge = { + isSystemUser = true; + group = "nar-bridge"; + }; + + users.groups.nar-bridge = { }; + + systemd.sockets.nar-bridge = { + description = "nar-bridge socket"; + wantedBy = [ "sockets.target" ]; + + socketConfig = { + LimitNOFILE = 65535; + ListenStream = "/run/nar-bridge.sock"; + SocketMode = "0666"; + SocketUser = "root"; + }; + }; + + systemd.services.nar-bridge = { + description = "NAR Bridge"; + requires = [ "nar-bridge.socket" ]; + after = [ "nar-bridge.socket" ]; + wantedBy = [ "multi-user.target" ]; + serviceConfig = { + ExecStart = "${package}/bin/nar-bridge ${lib.escapeShellArgs args}"; + + Restart = "always"; + RestartSec = "10"; + + User = "nar-bridge"; + Group = "nar-bridge"; + StateDirectory = "nar-bridge"; + }; + }; + }; +} diff --git a/users/flokli/nixos/nixos-tvix-cache/nar-bridge.nix b/users/flokli/nixos/nixos-tvix-cache/nar-bridge.nix new file mode 100644 index 000000000000..e87189da2a89 --- /dev/null +++ b/users/flokli/nixos/nixos-tvix-cache/nar-bridge.nix @@ -0,0 +1,141 @@ +{ config, depot, pkgs, ... }: +{ + imports = [ ./nar-bridge-module.nix ]; + + # Microbenchmark + # hyperfine --warmup 1 'rm -rf /tmp/cache; nix copy --from https://nixos.tvix.store/ --to "file:///tmp/cache?compression=none" /nix/store/jlkypcf54nrh4n6r0l62ryx93z752hb2-firefox-132.0' + # From a different hetzner machine with 1Gbps uplink: + # - with zstd: 13.384s + # - with gzip: 11.130s + # - with brotli: ~18s + # - without compression: 15.6s + + # From a 1Gbit link in TUM: + # - with zstd: 32.292s + # - with gzip: 51s + # - cache.nixos.org from the same connection: 36.559s + services.nginx = { + package = pkgs.nginxStable.override { + modules = [ pkgs.nginxModules.zstd ]; + }; + virtualHosts.${config.machine.domain} = { + # when using http2 we actually see worse throughput, + # because it only uses a single tcp connection, + # which pins nginx to a single core. + http2 = false; + locations."=/" = { + tryFiles = "$uri $uri/index.html =404"; + root = pkgs.runCommand "index" + { + nativeBuildInputs = [ depot.tools.cheddar ]; + } '' + mkdir -p $out + cheddar README.md < ${./README.md} > $out/index.html + find $out + ''; + }; + locations."/" = { + proxyPass = "http://unix:/run/nar-bridge.sock:/"; + extraConfig = '' + # Restrict allowed HTTP methods + limit_except GET HEAD { + # nar bridge allows to upload nars via PUT + deny all; + } + # Enable proxy cache + proxy_cache nar-bridge; + proxy_cache_key "$scheme$proxy_host$request_uri"; + proxy_cache_valid 200 301 302 10m; # Cache responses for 10 minutes + proxy_cache_valid 404 1m; # Cache 404 responses for 1 minute + proxy_cache_min_uses 2; # Cache only if the object is requested at least twice + proxy_cache_use_stale error timeout updating; + + zstd on; + zstd_types application/x-nix-nar; + ''; + }; + }; + + # use more cores for compression + appendConfig = '' + worker_processes auto; + ''; + + proxyCachePath."nar-bridge" = { + enable = true; + levels = "1:2"; + keysZoneName = "nar-bridge"; + # Put our 1TB NVME to good use + maxSize = "200G"; + inactive = "10d"; + useTempPath = false; + }; + }; + + services.nar-bridge = { + enable = true; + + settings = { + blobservices = { + root = { + type = "objectstore"; + object_store_url = "file:///var/lib/nar-bridge/blobs.object_store"; + object_store_options = { }; + }; + }; + + directoryservices = { + root = { + type = "redb"; + is_temporary = false; + path = "/var/lib/nar-bridge/directories.redb"; + }; + }; + + pathinfoservices = { + root = { + type = "cache"; + near = "redb"; + far = "cache-nixos-org"; + }; + + redb = { + type = "redb"; + is_temporary = false; + path = "/var/lib/nar-bridge/pathinfo.redb"; + }; + + "cache-nixos-org" = { + type = "nix"; + base_url = "https://cache.nixos.org"; + blob_service = "root"; + directory_service = "root"; + public_keys = [ + "cache.nixos.org-1:6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY=" + ]; + }; + }; + }; + }; + + systemd.tmpfiles.rules = [ + # Put the data in the big disk + "d /tank/nar-bridge 0755 nar-bridge nar-bridge -" + # Cache responses on NVME + "d /var/cache/nginx 0755 ${config.services.nginx.user} ${config.services.nginx.group} -" + ]; + + fileSystems."/var/lib/nar-bridge" = { + device = "/tank/nar-bridge"; + options = [ + "bind" + "nofail" + ]; + }; + + systemd.services.nar-bridge = { + unitConfig.RequiresMountsFor = "/var/lib/nar-bridge"; + # twice the normal allowed limit, same as nix-daemon + serviceConfig.LimitNOFILE = "1048576"; + }; +} |