Merge pull request #8157 from MinaProtocol/devnet2-and-mainnet-master

devnet-2 and mainnet

.gitignore

@@ -73,4 +73,6 @@ automation/block_tree.gv.png
 automation/gcloud-keyfile.json
 automation/services/watchdog/check_libp2p/check_libp2p
 
-*.backup
+*.backup
+
+*.terraform.lock.hcl

automation/bake/Dockerfile

@@ -24,8 +24,6 @@ RUN mkdir -p /root/.mina-config && echo "{}" > /root/.mina-config/daemon.json
 RUN mina daemon -config-file ${CONFIG_FILE} -generate-genesis-proof true; \
   mv ~/.mina-config/genesis/genesis_* /var/lib/coda/
 
-RUN curl https://raw.githubusercontent.com/MinaProtocol/mina/${COMMIT_HASH}/automation/terraform/testnets/${TESTNET_NAME}/peers.txt -o /root/peers.txt
-
 #RUN curl https://raw.githubusercontent.com/MinaProtocol/mina/develop/dockerfiles/scripts/healthcheck-utilities.sh -o /healthcheck/utilities.sh
 
 RUN echo '#!/bin/bash -x\n\
@@ -38,11 +36,11 @@ shift \n\
 while true; do\n\
   rm -f /root/.mina-config/.mina-lock\n\
   mina "$command" -config-file "'${CONFIG_FILE}'" "$@" 2>&1 >mina.log &\n\
-  coda_pid=$!\n\
+  mina_pid=$!\n\
   tail -q -f mina.log &\n\
   tail_pid=$!\n\
-  wait "$coda_pid"\n\
-  echo "Coda process exited with status code $?"\n\
+  wait "$mina_pid"\n\
+  echo "Mina process exited with status code $?"\n\
   sleep 10\n\
   kill "$tail_pid"\n\
   if [ ! -f stay_alive ]; then\n\

automation/scripts/patch-archive-node.sh

@@ -0,0 +1,27 @@
+#!/bin/bash
+
+
+IMAGE="codaprotocol/coda-archive:1.0.2"
+
+
+read -r -d '' PATCH << SPEC
+spec:
+  template:
+    spec:
+      containers:
+        - name: archive
+          image: $IMAGE
+          args:
+          - coda-archive
+          - run
+          - -postgres-uri
+          - postgres://postgres:foobar@archive-1-postgresql:5432/archive
+          - -config-file
+          - /config/daemon.json
+          - -server-port
+          - "3086"
+SPEC
+
+echo -e "$PATCH"
+
+kubectl patch deploy/archive-1 -p "$PATCH"

automation/scripts/patch-watchdog.sh

@@ -0,0 +1,37 @@
+#!/bin/bash
+
+
+IMAGE="gcr.io/o1labs-192920/watchdog:0.4.2"
+
+
+read -r -d '' PATCH << SPEC
+spec:
+  template:
+    spec:
+      containers:
+        - name: watchdog
+          image: $IMAGE
+SPEC
+
+echo -e "$PATCH"
+
+kubectl patch deploy/archive-1 -p "$PATCH"
+
+# ====================================
+
+
+read -r -d '' PATCH << SPEC
+spec:
+  jobTemplate:
+    spec:
+      template:
+        spec:
+          containers:
+            - name: watchdog
+              image: $IMAGE
+SPEC
+
+echo -e "$PATCH"
+
+kubectl patch cronjobs/watchdog-make-reports -p "$PATCH"
+

automation/services/watchdog/make_report.py

@@ -46,7 +46,7 @@ def main():
     parser.add_argument("-d", "--discord_webhook_url", help="discord webhook url", required=False, type=str, dest="discord_webhook_url")
     parser.add_argument("-a", "--accounts", help="community accounts csv", required=False, type=str, dest="accounts_csv")
     parser.add_argument("-l", "--local", help="run with a local node", required=False, type=bool, default=False)
-    parser.add_argument("-c", "--coda", help="local coda binary", required=False, type=str, default="coda")
+    parser.add_argument("-b", "--bin", help="local mina binary", required=False, type=str, default="mina", dest="binary")
 
     # ==========================================
 
@@ -85,7 +85,7 @@ def main():
     request_timeout_seconds = 600
 
     def exec_locally(command):
-      command = command.replace('coda', args.coda)
+      command = command.replace('mina', args.binary)
       print(command)
       import subprocess
       subprocess = subprocess.Popen(command, shell=True, stdout=subprocess.PIPE, text=True)

automation/services/watchdog/metrics.py

@@ -58,7 +58,6 @@ def pods_with_no_new_logs(v1, namespace, nodes_with_no_new_logs):
   for pod in pods.items:
     containers = pod.status.container_statuses
     mina_containers = list(filter(lambda c: c.name in [ 'coda', 'seed', 'coordinator' ], containers))
-    print(mina_containers)
     if len(mina_containers) != 0:
       name = pod.metadata.name
       recent_logs = v1.read_namespaced_pod_log(name=name, namespace=namespace, since_seconds=ten_minutes)

automation/services/watchdog/node_status_metrics.py

@@ -74,7 +74,7 @@ def get_deepest_child(p):
     return max(children_and_depths, key=lambda x: x[1])
 
   best_protocol_states = [ get_deepest_child(root) for root in roots ]
-  best_protocol_state = max(best_protocol_states, key=lambda x: x[1])[0]
+  best_protocol_state = max(best_protocol_states, key=lambda x: x[1], default=[ None ])[0]
 
   n = 3
   last_n_protocol_states = [ best_protocol_state ]
@@ -121,25 +121,26 @@ def add_resp(resp, direct_queried_peers):
         peer_table[k] = v
 
     queried_peers.update([ p['node_peer_id'] for p in peers ])
-    queried_peers.update([p['peer_id'] for p in direct_queried_peers])
+    queried_peers.update([ p['peer_id'] for p in direct_queried_peers ])
     for p in itertools.chain(*[ p['peers'] for p in peers ]):
-        unqueried_peers[p['peer_id']] = p
+      unqueried_peers[p['peer_id']] = p
     for p in queried_peers:
+      if p in unqueried_peers:
         del unqueried_peers[p]
 
   cmd = "mina advanced node-status -daemon-port " + seed_daemon_port + " -daemon-peers" + " -show-errors"
-  resp = util.exec_on_pod(v1, namespace, seed, 'seed', cmd)
+  resp = util.exec_on_pod(v1, namespace, seed, 'coda', cmd)
   add_resp(resp, [])
 
   requests = 0
 
   while len(unqueried_peers) > 0 and requests < max_crawl_requests:
     peers_to_query = list(unqueried_peers.values())
-    peers = ','.join(to_multiaddr_string(p) for p in peers_to_query)
+    peers = ','.join(peer_to_multiaddr(p) for p in peers_to_query)
 
     print ('Queried ' + str(len(queried_peers)) + ' peers. Gathering node status on %s unqueried peers'%(str(len(unqueried_peers))))
 
-    resp = util.exec_on_pod(v1, namespace, seed, 'seed', "mina advanced node-status -daemon-port " + seed_daemon_port + " -peers " + peers + " -show-errors")
+    resp = util.exec_on_pod(v1, namespace, seed, 'coda', "mina advanced node-status -daemon-port " + seed_daemon_port + " -peers " + peers + " -show-errors")
     add_resp(resp, peers_to_query)
 
     requests += 1

automation/services/watchdog/version.txt

@@ -1 +1 @@
-0.3.9
+0.4.3

automation/terraform/modules/kubernetes/testnet/helm.tf

@@ -171,7 +171,7 @@ resource "helm_release" "seeds" {
   name       = "${var.testnet_name}-seeds"
   repository = var.use_local_charts ? "" : local.mina_helm_repo
   chart      = var.use_local_charts ? "../../../../helm/seed-node" : "seed-node"
-  version    = "0.6.2"
+  version    = "1.0.3"
   namespace  = kubernetes_namespace.testnet_namespace.metadata[0].name
   values = [
     yamlencode(local.seed_vars)

automation/terraform/monitoring/o1-testnet-alerts.tf

@@ -14,7 +14,7 @@ module "o1testnet_alerts" {
 
   rule_filter            = "{testnet=~\".+\"}" # any non-empty testnet name
   rule_timeframe         = "1h"
-  pagerduty_alert_filter = "devnet"
+  pagerduty_alert_filter = "devnet|finalfinal2|mainnet|devnet2"
 }
 
 output "testnet_alert_rules" {