Merge #2956

2956: workbench: block propagation analysis improvements r=deepfire a=deepfire In the workbench / log analysis: 1. Finalise block propagation analysis 2. Improve concurrency 3. Improve timeline rendering 4. Fix issues Co-authored-by: Kosyrev Serge <serge.kosyrev@iohk.io>
IntersectMBO · Jul 27, 2021 · 7e999e4 · 7e999e4
2 parents f69205c + 92dec64
commit 7e999e4
Show file tree

Hide file tree

Showing 19 changed files with 737 additions and 539 deletions.
diff --git a/Makefile b/Makefile
@@ -43,14 +43,12 @@ CLUSTER_PROFILE    ?= default-alzo
 CLUSTER_ARGS_EXTRA ?=
 
 cluster-shell:
-	nix-shell --max-jobs 8 --cores 0 --argstr clusterProfile ${CLUSTER_PROFILE} --arg 'autoStartCluster' true ${CLUSTER_ARGS_EXTRA}
-
-cluster-shell-dev:               CLUSTER_ARGS_EXTRA += --arg 'workbenchDevMode' true
-cluster-shell-trace:             CLUSTER_ARGS_EXTRA += --trace
-large-state-cluster-shell-trace: CLUSTER_ARGS_EXTRA += --trace
-large-state-cluster-shell:       CLUSTER_PROFILE = k2-10ep-2000kU-500kD-nobs-mary
-large-state-cluster-shell-trace: CLUSTER_PROFILE = k2-10ep-2000kU-500kD-nobs-mary
-cluster-shell-dev cluster-shell-trace large-state-cluster-shell large-state-cluster-shell-trace: cluster-shell
+	nix-shell --max-jobs 8 --cores 0 --show-trace --argstr clusterProfile ${CLUSTER_PROFILE} --arg 'autoStartCluster' true ${CLUSTER_ARGS_EXTRA}
+
+cluster-shell-dev:       CLUSTER_ARGS_EXTRA += --arg 'workbenchDevMode' true
+cluster-shell-trace:     CLUSTER_ARGS_EXTRA += --argstr 'autoStartClusterArgs' '--trace --trace-workbench'
+cluster-shell-dev-trace: CLUSTER_ARGS_EXTRA += --arg 'workbenchDevMode' true --argstr 'autoStartClusterArgs' '--trace --trace-workbench'
+cluster-shell-dev cluster-shell-trace cluster-shell-dev-trace: cluster-shell
 
 cli node:
 	cabal --ghc-options="+RTS -qn8 -A32M -RTS" build cardano-$@

diff --git a/nix/supervisord-cluster/default.nix b/nix/supervisord-cluster/default.nix
@@ -172,54 +172,64 @@ EOF
     export PATH=$PATH:${path}
 
     wb backend assert-is supervisor
-    wb backend pre-run-hook    "${stateDir}"
+    wb backend assert-stopped
 
     wb_run_allocate_args=(
         --cache-dir            "${cacheDir}"
         --base-port             ${toString basePort}
         --stagger-ports
+        --
         --port-shift-ekg        100
         --port-shift-prometheus 200
+        --supervisor-conf      "${backend.supervisord.mkSupervisorConf profile}"
       )
     wb run allocate $batch_name ${profile.name} "''${wb_run_allocate_args[@]}"
-    rm -f                      "${stateDir}"
-    ln -sf         run/current "${stateDir}"
 
     current_run_path=$(wb run current-path)
     ${workbench.initialiseProfileRunDirShellScript profile "$current_run_path"}
 
-    wb_run_start_args=(
-        --supervisor-conf      "${backend.supervisord.mkSupervisorConf profile}"
-      )
-    wb run start "''${run_start_flags[@]}" $(wb run current-tag) "''${wb_run_start_args[@]}"
+    wb run start "''${run_start_flags[@]}" $(wb run current-tag)
 
     echo 'workbench:  cluster started. Run `stop-cluster` to stop'
   '';
+
   stop = pkgs.writeScriptBin "stop-cluster" ''
     set -euo pipefail
 
     while test $# -gt 0
     do case "$1" in
-        --trace | --debug ) set -x;;
+        --trace | --debug )              set -x;;
+        --trace-wb | --trace-workbench ) export WORKBENCH_EXTRA_FLAGS=--trace;;
+        * ) break;; esac; shift; done
+
+    wb run stop $(wb run current-tag)
+  '';
+
+  restart = pkgs.writeScriptBin "restart-cluster" ''
+    set -euo pipefail
+
+    wb_flags=()
+    wb_run_restart_flags=()
+
+    while test $# -gt 0
+    do case "$1" in
+        --no-generator | --no-gen )      wb_run_restart_flags+=($1);;
+        --trace | --debug | --trace-wb | --trace-workbench )
+                                         wb_flags+=(--trace);;
+        --help )                         cat <<EOF
+${startClusterUsage}
+EOF
+                                         exit 1;;
         * ) break;; esac; shift; done
 
-    ${pkgs.python3Packages.supervisor}/bin/supervisorctl stop all
-    if wb backend is-running
-    then
-      if test -f "${stateDir}/supervisor/cardano-node.pids"
-      then kill $(<${stateDir}/supervisor/supervisord.pid) $(<${stateDir}/supervisor/cardano-node.pids)
-      else pkill supervisord
-      fi
-      echo "workbench:  cluster terminated"
-      rm -f ${stateDir}/supervisor/supervisord.pid ${stateDir}/supervisor/cardano-node.pids
-    else
-      echo "workbench:  cluster is not running"
-    fi
+    wb "''${wb_flags[@]}" run restart "''${wb_run_restart_flags[@]}"
+
+    echo "workbench:  alternate command for this action:  wb run restart"
   '';
 
 in
 {
   inherit workbench;
   inherit (workbenchProfiles) profilesJSON;
-  inherit profile stateDir start stop;
+  inherit profile stateDir start stop restart;
 }
diff --git a/nix/workbench/analyse.sh b/nix/workbench/analyse.sh
@@ -9,26 +9,24 @@ usage_analyse() {
     Options of 'analyse' command:
 
        --reanalyse        Skip the preparatory steps and launch 'locli' directly
-       --time             Time the 'locli' executable runs
 EOF
 }
 
 analyse() {
-local skip_preparation= time= dump_logobjects=
+local skip_preparation= time= dump_logobjects= args=()
 while test $# -gt 0
 do case "$1" in
-       --reanalyse | --re ) skip_preparation='true';;
-       --dump-logobjects )  dump_logobjects='true';;
-       --time )             time='eval time';;
+       --reanalyse | --re ) skip_preparation='true'; args+=($1);;
+       --dump-logobjects )  dump_logobjects='true';  args+=($1);;
        * ) break;; esac; shift; done
 
 local op=${1:-$(usage_analyse)}; shift
 
 case "$op" in
     block-propagation | bp )
-        local usage="USAGE: wb analyse $op [RUN-NAME=current]"
+        local usage="USAGE: wb analyse $op [RUN-NAME=current].."
 
-        local name=${1:-current}
+        local name=${1:-current}; shift
         local dir=$(run get "$name")
         local adir=$dir/analysis
         if test -z "$dir"
@@ -82,8 +80,11 @@ case "$op" in
         if test -n "$dump_logobjects"; then
             locli_args+=(--logobjects-json "$adir"/logs-cluster.logobjects.json); fi
 
-        ${time} locli 'analyse' 'block-propagation' \
-            "${locli_args[@]}" "$adir"/*.flt.json;;
+        time locli 'analyse' 'block-propagation' \
+             "${locli_args[@]}" "$adir"/*.flt.json
+
+        ## More than one run passed?
+        test $# -gt 0 && analyse ${args[*]} block-propagation "$@";;
 
     grep-filtered-logs | grep | g )
         local usage="USAGE: wb analyse $op BLOCK [MACHSPEC=*] [RUN-NAME=current]"
@@ -132,7 +133,7 @@ case "$op" in
         locli analyse substring-keys | grep -v 'Temporary modify' > "$keyfile"
 
         if test "$mach" = 'all'
-        then local machs=($(wb run list-hosts $name))
+        then local machs=($(run list-hosts $name))
         else local machs=($mach); fi
 
         for mach in ${machs[*]}
@@ -164,8 +165,8 @@ case "$op" in
            if test -n "$dump_logobjects"; then
                locli_args+=(--logobjects-json "$adir"/logs-$mach.logobjects.json); fi
 
-           ${time} locli 'analyse' 'machine-timeline' \
-                   "${locli_args[@]}" "$consolidated"
+           time locli 'analyse' 'machine-timeline' \
+                "${locli_args[@]}" "$consolidated"
            ) &
         done
 

diff --git a/nix/workbench/backend.sh b/nix/workbench/backend.sh
@@ -1,6 +1,8 @@
 usage_backend() {
      usage "backend" "Abstract over cluster backend operations" <<EOF
-    is-running       Test if cluster is running
+    is-running RUNDIR
+                     Test if the cluster specified by the run directory
+                       is currently running
 
     get-node-socket-path RUNDIR
                      Given a run directory, print the node socket path
@@ -10,15 +12,18 @@ usage_backend() {
                      Extend the environment JSON file with backend-specific
                        environment config
 
+    allocate-run RUNDIR
     describe-run RUNDIR
-    pre-run-hook RUNDIR
-    lostream-fixup-jqargs RUNDIR
-    lostream-fixup-jqexpr
-
     start-cluster RUNDIR
                      Start the cluster nodes
     start-generator RUNDIR
                      Start generator
+    cleanup-cluster RUNDIR
+                     Wipe cluster state to pristine
+    stop-cluster RUNDIR
+
+    lostream-fixup-jqargs RUNDIR
+    lostream-fixup-jqexpr
 
     assert-is BACKEND-NAME
                      Check that the current backend is as expected
@@ -31,27 +36,29 @@ backend() {
 local op=${1:-$(usage_backend)} # No need to shift -- backends will use the op.
 
 case "${op}" in
-    is-running )                 $WORKBENCH_BACKEND "$@";;
-    get-node-socket-path )       $WORKBENCH_BACKEND "$@";;
-    record-extended-env-config ) $WORKBENCH_BACKEND "$@";;
-    describe-run )               $WORKBENCH_BACKEND "$@";;
-    pre-run-hook )               $WORKBENCH_BACKEND "$@";;
-    start-cluster )              $WORKBENCH_BACKEND "$@";;
-    start-generator )            $WORKBENCH_BACKEND "$@";;
-    lostream-fixup-jqargs )      $WORKBENCH_BACKEND "$@";;
-    lostream-fixup-jqexpr )      $WORKBENCH_BACKEND "$@";;
+    is-running )                 backend_$WORKBENCH_BACKEND "$@";;
+    get-node-socket-path )       backend_$WORKBENCH_BACKEND "$@";;
+    record-extended-env-config ) backend_$WORKBENCH_BACKEND "$@";;
+    allocate-run )               backend_$WORKBENCH_BACKEND "$@";;
+    describe-run )               backend_$WORKBENCH_BACKEND "$@";;
+    start-cluster )              backend_$WORKBENCH_BACKEND "$@";;
+    start-generator )            backend_$WORKBENCH_BACKEND "$@";;
+    cleanup-cluster )            backend_$WORKBENCH_BACKEND "$@";;
+    stop-cluster )               backend_$WORKBENCH_BACKEND "$@";;
+    lostream-fixup-jqargs )      backend_$WORKBENCH_BACKEND "$@";;
+    lostream-fixup-jqexpr )      backend_$WORKBENCH_BACKEND "$@";;
 
     ## Handle non-generic calls:
-    passthrough | pass )         $WORKBENCH_BACKEND "$@";;
+    passthrough | pass )         backend_$WORKBENCH_BACKEND "$@";;
 
     assert-is )
         local usage="USAGE: wb run $op BACKEND-NAME"
         local name=${2:?$usage}
 
         ## Check the backend echoes own name:
-        local actual_name=$($WORKBENCH_BACKEND name)
+        local actual_name=$(backend_$WORKBENCH_BACKEND name)
         if test "$actual_name" != "$name"
-        then fatal "Workbench is broken:  '$WORKBENCH_BACKEND name' returned:  '$actual_name'"; fi
+        then fatal "Workbench is broken:  'workbench_$WORKBENCH_BACKEND name' returned:  '$actual_name'"; fi
         ;;
 
     assert-stopped )

diff --git a/nix/workbench/default.nix b/nix/workbench/default.nix
@@ -88,7 +88,7 @@ let
     else "nix-exes+checkout-wb";
 
   shellHook = ''
-    export WORKBENCH_BACKEND=${./.}/supervisor.sh
+    export WORKBENCH_BACKEND=supervisor
 
       ${optionalString workbenchDevMode
     ''

diff --git a/nix/workbench/lib.sh b/nix/workbench/lib.sh
@@ -58,7 +58,6 @@ git_repo_commit_description() {
     local commit=$(git -C "$repo" rev-parse 'HEAD' 2>/dev/null || true)
 
     test -n "$commit" && {
-        git -C "$repo" fetch
         git -C "$repo" describe --match '[0-9].*' --tags $commit 2>/dev/null |
             cut -d'-' -f1,2 | tr -d '\n'
         git -C "$repo" diff --exit-code --quiet || echo '-modified'

diff --git a/nix/workbench/locli/locli.cabal b/nix/workbench/locli/locli.cabal
@@ -20,6 +20,7 @@ library
                        Cardano.Analysis.Profile
                        Cardano.Analysis.TopHandler
 
+                       Cardano.Analysis.API
                        Cardano.Analysis.BlockProp
                        Cardano.Analysis.Driver
                        Cardano.Analysis.MachTimeline
@@ -44,6 +45,7 @@ library
                      , cardano-config
                      , cardano-prelude
                      , containers
+                     , deepseq
                      , directory
                      , filepath
                      , file-embed
@@ -54,6 +56,7 @@ library
                      , optparse-generic
                      , ouroboros-network
                      , process
+                     , quiet
                      , scientific
                      , split
                      , statistics
@@ -87,7 +90,7 @@ executable locli
   ghc-options:         -threaded
                        -Wall
                        -rtsopts
-                       "-with-rtsopts=-T"
+                       "-with-rtsopts=-T -N4 -A8m"
   build-depends:       base
                      , cardano-prelude
                      , locli