refactor, separate out check_bootnodes and make it posix-compliant

scripts/ci/common/lib.sh

@@ -139,3 +139,50 @@ has_runtime_changes() {
     return 1
   fi
 }
+
+# given a bootnode and the path to a chainspec file, this function will create a new chainspec file
+# with only the bootnode specified and test whether that bootnode provides peers
+# The optional third argument is the index of the bootnode in the list of bootnodes, this is just used to pick an ephemeral
+# port for the node to run on. If you're only testing one, it'll just use the first ephemeral port
+# BOOTNODE: /dns/polkadot-connect-0.parity.io/tcp/443/wss/p2p/12D3KooWEPmjoRpDSUuiTjvyNDd8fejZ9eNWH5bE965nyBMDrB4o
+# CHAINSPEC_FILE: /path/to/polkadot.json
+check_bootnode(){
+    BOOTNODE=$1
+    BASE_CHAINSPEC=$2
+    RUNTIME=$(basename "$BASE_CHAINSPEC" | cut -d '.' -f 1)
+
+    # Generate a temporary chainspec file containing only the bootnode we care about
+    TMP_CHAINSPEC_FILE="$RUNTIME.$(echo "$BOOTNODE" | tr '/' '_').tmp.json"
+    jq ".bootNodes = [\"$BOOTNODE\"] " < "$CHAINSPEC_FILE" > "$TMP_CHAINSPEC_FILE"
+
+    # Grab an unused port by binding to port 0 and then immediately closing the socket
+    # This is a bit of a hack, but it's the only way to do it in the shell
+    RPC_PORT=$(python -c "import socket; s=socket.socket(); s.bind(('', 0)); print(s.getsockname()[1]); s.close()")
+
+    echo "[+] Checking bootnode $BOOTNODE"
+    polkadot --chain "$TMP_CHAINSPEC_FILE" --no-mdns --rpc-port="$RPC_PORT" --tmp  > /dev/null 2>&1 &
+    POLKADOT_PID=$!
+    # Wait a little while for the node to start up and attempt to start peering
+    sleep 60
+    # Delete the temporary chainspec file now we're done running the node
+    rm "$TMP_CHAINSPEC_FILE"
+
+    # Check the health endpoint of the RPC node
+    PEERS="$(curl -s -X POST -H "Content-Type: application/json" --data '{"jsonrpc":"2.0","method":"system_health","params":[],"id":1}' http://localhost:"$RPC_PORT" | jq -r '.result.peers')"
+    # Clean up the node
+    kill -9 $POLKADOT_PID
+    # Sometimes due to machine load or other reasons, we don't get a response from the RPC node
+    # If $PEERS is an empty variable, mark the node as unreachable
+    if [ -z "$PEERS" ]; then
+        PEERS=0
+    fi
+    if [ "$PEERS" -gt 0 ]; then
+        echo "[+] $PEERS peers found for $BOOTNODE"
+        echo "    Bootnode appears contactable"
+        return 0
+    else
+        echo "[!] $PEERS peers found for $BOOTNODE"
+        echo "    Bootnode appears unreachable"
+        return 1
+    fi
+}

scripts/ci/github/check_bootnodes.sh

@@ -9,6 +9,8 @@
 
 # Root of the polkadot dir
 ROOT="$(dirname "${0}")/../../.."
+# shellcheck source=scripts/ci/common/lib.sh
+source "$ROOT/scripts/ci/common/lib.sh"
 RUNTIME="$1"
 
 trap cleanup EXIT INT TERM
@@ -19,67 +21,21 @@ cleanup(){
     killall polkadot > /dev/null 2>&1
 }
 
-check_bootnode(){
-    BOOTNODE_INDEX=$1
-    TMP_CHAINSPEC_FILE="$RUNTIME.$BOOTNODE_INDEX.tmp.json"
-    FINAL_CHAINSPEC_FILE="$RUNTIME.$BOOTNODE_INDEX.final.json"
-    # Copy the chainspec file to a temporary location to avoid weird race conditions when running in parallel
-    cp "$CHAINSPEC_FILE" "$CHAINSPEC_TMPDIR/$TMP_CHAINSPEC_FILE"
-    pushd "$CHAINSPEC_TMPDIR" > /dev/null || exit 1
-        jq ".bootNodes |= [.[$BOOTNODE_INDEX]] " < "$TMP_CHAINSPEC_FILE" > "$FINAL_CHAINSPEC_FILE"
-        BOOTNODE=$( jq -r '.bootNodes[0]' < "$FINAL_CHAINSPEC_FILE" )
-        # Get the first ephemeral port
-        if [[ "$OSTYPE" == "darwin"* ]]; then
-            BASE_PORT=$(sysctl net.inet.ip.portrange.first | awk '{print $2}')
-        else
-            BASE_PORT=$(sysctl net.ipv4.ip_local_port_range | awk '{print $3}')
-        fi
-        # If we're on a mac, use this instead
-        # BASE_PORT=$(sysctl net.inet.ip.portrange.first | awk '{print $2}')
-        RPC_PORT=$((BASE_PORT + BOOTNODE_INDEX))
-        echo "[+] Checking bootnode $BOOTNODE"
-        polkadot --chain "$FINAL_CHAINSPEC_FILE" --no-mdns --rpc-port=$RPC_PORT --tmp  > /dev/null 2>&1 &
-        POLKADOT_PID=$!
-        # We just spun up a bunch of nodes... probably want to wait a bit.
-        sleep 60
-    popd > /dev/null || exit 1
-    # Check the health endpoint of the RPC node
-    PEERS="$(curl -s -X POST -H "Content-Type: application/json" --data '{"jsonrpc":"2.0","method":"system_health","params":[],"id":1}' http://localhost:$RPC_PORT | jq -r '.result.peers')"
-    # Clean up the node
-    kill -9 $POLKADOT_PID
-    # Sometimes due to machine load or other reasons, we don't get a response from the RPC node
-    # If $PEERS is an empty variable, mark the node as unreachable
-    if [ -z "$PEERS" ]; then
-        PEERS=0
-    fi
-    if [ "$PEERS" -gt 0 ]; then
-        echo "[+] $PEERS peers found for $BOOTNODE"
-        echo "    Bootnode appears contactable"
-        GOOD_BOOTNODES+=("$BOOTNODE")
-        return 0
-    else
-        echo "[!] $PEERS peers found for $BOOTNODE"
-        echo "    Bootnode appears unreachable"
-        BAD_BOOTNODES+=("$BOOTNODE")
-        return 1
-    fi
-}
-
 # For each runtime
 CHAINSPEC_FILE="$ROOT/node/service/chain-specs/$RUNTIME.json"
 # count the number of bootnodes
 BOOTNODES=$( jq -r '.bootNodes | length' "$CHAINSPEC_FILE" )
 # Make a temporary dir for chainspec files
-CHAINSPEC_TMPDIR="$(mktemp -d -t "${RUNTIME}_chainspecs_XXXXXX")"
-echo "[+] Using $CHAINSPEC_TMPDIR as temporary chainspec dir"
 # Store an array of the bad bootnodes
 BAD_BOOTNODES=()
 GOOD_BOOTNODES=()
 PIDS=()
+
 echo "[+] Checking $BOOTNODES bootnodes for $RUNTIME"
 for i in $(seq 0 $((BOOTNODES-1))); do
+    BOOTNODE=$( jq -r .bootNodes["$i"] < "$CHAINSPEC_FILE" )
     # Check each bootnode in parallel
-    check_bootnode "$i" &
+    check_bootnode "$BOOTNODE" "$CHAINSPEC_FILE" &
     PIDS+=($!)
     # Hold off 10 seconds between attempting to spawn nodes to stop the machine from getting overloaded
     sleep 10