WIP CAD-2839 supervisord | ctl: cluster control tool

Makefile

@@ -34,16 +34,16 @@ test-chairmans-cluster:
 	@scripts/chairmans-cluster/cluster-test.sh
 
 profiles:
-	@jq .    $$(nix-build -A profiles)
+	@./nix/supervisord-cluster/ctl dump-profiles
 
 profile-names:
-	@jq keys $$(nix-build -A profiles)
+	@./nix/supervisord-cluster/ctl profile-names
 
 CLUSTER_PROFILE    = default-mary
 CLUSTER_ARGS_EXTRA =
 
 cluster-shell:
-	nix-shell --max-jobs 8 --cores 0 --command 'start-cluster; return' --argstr clusterProfile ${CLUSTER_PROFILE} --command 'start-cluster ${CLUSTER_ARGS_EXTRA}; return'
+	nix-shell --max-jobs 8 --cores 0 --arg 'autoStartCluster' true --argstr clusterProfile ${CLUSTER_PROFILE}
 
 cluster-shell-trace:             CLUSTER_ARGS_EXTRA = --trace
 large-state-cluster-shell-trace: CLUSTER_ARGS_EXTRA = --trace

nix/pkgs.nix

@@ -67,8 +67,7 @@ pkgs: _: with pkgs;
   inherit (cardanoNodeHaskellPackages.ouroboros-consensus-byron.components.exes) db-converter;
   inherit (cardanoNodeHaskellPackages.network-mux.components.exes) cardano-ping;
 
-  mkCluster = cfg: callPackage ./supervisord-cluster
-                     (lib.traceValSeqN 2 cfg);
+  mkCluster = cfg: callPackage ./supervisord-cluster cfg;
   cardanolib-py = callPackage ./cardanolib-py {};
 
   clusterTests = import ./supervisord-cluster/tests { inherit pkgs; };

nix/supervisord-cluster/ctl/ctl

@@ -0,0 +1,76 @@
+#!/usr/bin/env bash
+# shellcheck disable=SC1090,SC2046,SC2206,SC2207
+
+set -euo pipefail
+
+global_ctl_basedir=${global_ctl_basedir:-$(realpath "$(dirname "$0")")}
+global_ctl_runsdir=${global_ctl_runsdir:-$(realpath "$global_ctl_basedir/../../runs")}
+
+. "$global_ctl_basedir"/ctl-lib.sh
+. "$global_ctl_basedir"/ctl-profile.sh
+. "$global_ctl_basedir"/ctl-run.sh
+. "$global_ctl_basedir"/ctl-supervisor.sh
+. "$global_ctl_basedir"/ctl-topology.sh
+
+usage_main() {
+     __usage "OP" "Main OPs" <<EOF
+    profile    (p)        Benchmark profile operations.  Default subop is 'list'
+    run        (r)        Managing benchmark runs.  Default subop is 'list'
+    topology   (t)        Topology generation
+
+    supervisor (s)        Managing local cluster
+
+  Options:
+
+    --trace / --debug     set -x
+    --help                This short help.
+    --help-full           Extended help.
+EOF
+}
+
+usage_extra() {
+    cat >&2 <<EOF
+  Other OPs:
+
+    call ARGS..           Call internal functions with arguments.
+
+  Extra options:
+
+    --cls                 Clear screen, before acting further.
+    --runsdir DIR         Set the runs directory.  Defaults to $global_ctl_runsdir
+
+EOF
+}
+
+while test $# -gt 0
+do case "$1" in
+       --runsdir )
+           global_ctl_runsdir=$2; shift;;
+       --cls )
+           echo -en "\ec">&2;;
+       --trace | --debug )
+           set -x;;
+       --help )
+           usage_main; exit 1;;
+       --help-full | --help-all | --help-extra )
+           usage_main; usage_extra; exit 1;;
+       * ) break;; esac; shift; done
+
+main() {
+    local op=${1:?$(usage_main)}; shift
+
+    case "${op}" in
+        profile | profiles | prof | ps | p )
+            profile "$@";;
+        supervisor | sup | sv | s )
+            supervisor "$@";;
+        topology | topo | t )
+            topology "$@";;
+        run | runs | r )
+            run "$@";;
+        call )
+            "$@";;
+        * ) usage_main; exit 1;; esac
+}
+
+main "$@"

nix/supervisord-cluster/ctl/ctl-lib.sh

@@ -0,0 +1,31 @@
+to_jsonlist() {
+    for x in "$@"
+    do echo "\"$x\""
+    done | jq --slurp '.'
+}
+
+__usage() {
+    local op=$1 desc=$2
+    cat >&2 <<EOF
+USAGE:  $(basename "$0") OPTIONS.. $op SUBOP SUBOP-ARGS..
+
+  $desc:
+
+$(cat)
+
+EOF
+}
+
+usage() {
+    __usage "$@"
+    exit 1
+}
+
+fail() {
+    echo "$*" >&2
+    exit 1
+}
+
+fatal() {
+    fail "FATAL: $*"
+}

nix/supervisord-cluster/ctl/ctl-profile.sh

@@ -0,0 +1,119 @@
+usage_profile() {
+     usage "profile" "Benchmark profile operations" <<EOF
+    list                  List profile names (json)
+    all-profiles | all    All profile contents (json)
+    compose NAME..        Create a profile composed from named profiles
+    get NAME              Get contents of named profile
+    describe NAME         Print a human description of a profile
+    node-specs NAME PORT-BASE STAGGER-PORTS
+                          Print node specs JSON for the given profile;
+                            If STAGGER-PORTS is true, the assigned ports will
+                            be incrementally distinc for each node spec
+EOF
+}
+
+global_profile_eras=(
+    shelley
+    allegra
+    mary
+)
+
+profile() {
+local op=${1:-list}; test $# -gt 0 && shift
+
+case "${op}" in
+    list | names )
+        profile generate-all | jq 'keys'
+        ;;
+
+    all-profiles | generate-all | all )
+        (cd "$global_ctl_basedir";
+
+         jq --argjson eras "$(to_jsonlist ${global_profile_eras[*]})" --null-input '
+           include "profiles";
+
+           $eras
+           | map(profiles(.; null; null; []))
+           | add
+           ');;
+
+    compose )
+        local profile_names="$@"
+
+        profile generate-all |
+        jq --argjson profile_names "$(to_jsonlist ${profile_names[*]})" '
+          . as $profiles
+          | $profile_names | debug
+          | map($profiles[.])
+          | add
+          ';;
+
+    get )
+        local usage="USAGE: ctl profile get NAME"
+        local name=${1:?$usage}
+
+        profile generate-all |
+        jq '.["'$name'"]'
+        ;;
+
+    describe )
+        local usage="USAGE: ctl profile describe NAME"
+        local name=${1:?$usage}
+
+        profile get $name |
+        (cd "$global_ctl_basedir";
+
+         jq '
+          include "derived";
+          profile_pretty_describe(.)
+          ' --raw-output);;
+
+    node-specs )
+        local usage="USAGE: ctl profile node-specs NAME PORT-BASE STAGGER-PORTS"
+        local name=${1:?$usage}
+        local port_base=${2:?$usage}
+        local stagger_ports=${3:?$usage}
+
+        args=(
+            --argjson port_base     $port_base
+            --argjson stagger_ports $stagger_ports
+        )
+        profile get $name |
+        jq '. as $prof
+           | $prof.composition.n_bft_hosts  as $n_bfts
+           | $prof.composition.n_pool_hosts as $n_pools
+           | ([range(0;
+                     $n_bfts)]
+              | map({ i: .
+                    , kind: "bft"
+                    }))
+              as $bfts
+           | ([range($n_bfts;
+                     $n_bfts + $n_pools)]
+              | map({ i: .
+                    , kind: "pool"
+                    }))
+              as $pools
+           | ([range($n_bfts + $n_pools;
+                     $n_bfts + $n_pools +
+                     if $prof.composition.with_observer then 1 else 0 end)]
+              | map({ i: .
+                    , kind: "observer"
+                    }))
+              as $observers
+           | ($bfts + $pools + $observers
+              | map(. +
+                    { name:       "node-\(.["i"])"
+                    , isProducer: ([.kind == "bft", .kind == "pool"] | any)
+                    , port:
+                      (if $stagger_ports
+                       then $port_base + .i
+                       else $port_base
+                       end)
+                    }))
+           | map({ key: .name, value: .})
+           | from_entries
+           ' "${args[@]}";;
+
+    * ) usage_profile;; esac
+}

nix/supervisord-cluster/ctl/ctl-run.sh

@@ -0,0 +1,37 @@
+usage_run() {
+     usage "run" "Managing benchmark runs" <<EOF
+    list                  List benchmark runs
+    mkname                Allocate a benchmark run name/id/tag
+    create                Create a benchmark run storage directory
+EOF
+}
+
+run() {
+local op=${1:-list}; test $# -gt 0 && shift
+
+case "${op}" in
+    list )
+        (test -d "$global_ctl_runsdir" && cd "$global_ctl_runsdir" && find . -type d)
+        ;;
+
+    mkname )
+        local batch=$1
+        local profjson=$2
+        echo "$(date +'%Y'-'%m'-'%d'-'%H.%M').$batch.$prof"
+        ;;
+
+    create )
+        ## Assumptions:
+        ##   - genesis has been created
+        ##   - the cluster is operating
+        local name=$1
+        local batch=$2
+        local prof=$3
+
+        local dir=$runsdir/$name
+        if test "$(realpath "$dir")" = test "$(realpath "$global_ctl_runsdir")" -o "$name" = '.'
+        then fatal "bad, bad tag '$name'"; fi
+        ;;
+
+    * ) usage_run;; esac
+}

nix/supervisord-cluster/ctl/ctl-supervisor.sh

@@ -0,0 +1,19 @@
+usage_supervisor() {
+     usage "supervisor" "Managing local cluster" <<EOF
+    assert-stopped   Assert that 'supervisord' is not running
+    is-running       Test if 'supervisord' is running
+EOF
+}
+
+supervisor() {
+local op=${1:-$(usage_supervisor)}; shift
+
+case "${op}" in
+    assert-stopped )
+        supervisor is-running &&
+          fail "Supervisord is already running. Please run 'stop-cluster' first!" ||
+          true;;
+    is-running )
+        test "$(netstat -pltn 2>/dev/null | grep ':9001 ' | wc -l)" != "0";;
+    * ) usage_supervisor;; esac
+}