fix error

checker.py

@@ -0,0 +1,168 @@
+# flake8: noqa
+# pylint: disable=broad-exception-raised, raise-missing-from, too-many-arguments, redefined-outer-name
+# pylance: disable=reportMissingImports, reportMissingModuleSource, reportGeneralTypeIssues
+# type: ignore
+
+import re
+from datetime import datetime, timedelta, timezone
+import subprocess
+import os
+import time
+import random
+import sys
+import pkg_resources
+import requests
+import json
+required_packages = ['grist-api', 'colorama']
+installed_packages = [pkg.key for pkg in pkg_resources.working_set]
+
+for package in required_packages:
+    if package not in installed_packages:
+        subprocess.check_call([sys.executable, '-m', 'pip', 'install', package, '--break-system-packages'])
+
+from grist_api import GristDocAPI
+import colorama
+
+import logging
+import socket
+
+
+class GRIST:
+    def __init__(self, server, doc_id, api_key, logger):
+        self.server = server
+        self.doc_id = doc_id
+        self.api_key = api_key
+        self.logger = logger
+        self.grist = GristDocAPI(doc_id, server=server, api_key=api_key)
+
+    def table_name_convert(self, table_name):
+        return table_name.replace(" ", "_")
+
+    def to_timestamp(self, dtime: datetime) -> int:
+        if dtime.tzinfo is None:
+            dtime = dtime.replace(tzinfo=timezone(timedelta(hours=3))) 
+        return int(dtime.timestamp())
+
+    def insert_row(self, data, table):
+        data = {key.replace(" ", "_"): value for key, value in data.items()}
+        row_id = self.grist.add_records(self.table_name_convert(table), [data])
+        return row_id
+
+    def update_column(self, row_id, column_name, value, table):
+        if isinstance(value, datetime):
+            value = self.to_timestamp(value)
+        column_name = column_name.replace(" ", "_")
+        self.grist.update_records(self.table_name_convert(table), [{ "id": row_id, column_name: value }])
+
+    def delete_row(self, row_id, table):
+        self.grist.delete_records(self.table_name_convert(table), [row_id])
+
+    def update(self, row_id, updates, table):
+        for column_name, value in updates.items():
+            if isinstance(value, datetime):
+                updates[column_name] = self.to_timestamp(value)
+        updates = {column_name.replace(" ", "_"): value for column_name, value in updates.items()}
+        self.grist.update_records(self.table_name_convert(table), [{"id": row_id, **updates}])
+
+    def fetch_table(self, table):
+        return self.grist.fetch_table(self.table_name_convert(table))
+
+    def find_record(self, id=None, state=None, name=None, table=None):
+        if table is None:
+            raise ValueError("Table is not specified")
+        table_data = self.grist.fetch_table(self.table_name_convert(table))
+        if id is not None:
+            record = [row for row in table_data if row.id == id]
+            return record
+        if state is not None and name is not None:
+            record = [row for row in table_data if row.State == state and row.name == name]
+            return record
+        if state is not None:
+            record = [row for row in table_data if row.State == state]
+            return record
+        if name is not None:
+            record = [row for row in table_data if row.Name == name]
+            return record
+
+    def find_settings(self, key, table):
+        table = self.fetch_table(self.table_name_convert(table))
+        for record in table:
+            if record.Setting == key:
+                if record.Value is None or record.Value == "":
+                    raise ValueError(f"Setting {key} blank")
+                return record.Value
+        raise ValueError(f"Setting {key} not found")
+
+
+def check_logs(log_handler):
+    # Initialize counters
+    error_count = 0
+    sync_count = 0
+    total_challenges = 0
+
+    try:
+        logs = subprocess.run(['docker', 'compose', 'logs', '--since', '24h'], cwd='/root/node/', capture_output=True, text=True, check=True)
+        log_content = logs.stdout
+    except subprocess.CalledProcessError as e:
+        raise RuntimeError(f"Error running docker compose logs: {e}") from e
+    
+    for line in log_content.split('\n'):
+        if "Error from tendermint rpc" in line: 
+            error_count += 1
+            log_handler.error(f"RPC error: {line}")
+        if "Is your verifier's account funded" in line:
+            log_handler.error(f"Verifier account not funded: {line}")
+            error_count += 1
+        if "Synced with network" in line:
+            sync_count += 1
+            log_handler.info(f"Synced with network: {line}")
+
+        challenge_match = re.search(r'made (\d+) secret challenges', line)
+        if challenge_match:
+            total_challenges += int(challenge_match.group(1))
+            log_handler.info(f"Made {total_challenges} secret challenges: {line}")
+    data = {
+        "errors": error_count,
+        "sync_events": sync_count, 
+        "total_challenges": total_challenges
+    }
+    log_handler.info(f"Result: {data}")
+    return data
+
+if __name__ == "__main__":
+    colorama.init(autoreset=True)
+    logger = logging.getLogger("Checker")
+    logger.setLevel(logging.INFO)
+    formatter = logging.Formatter("%(asctime)s - %(name)s - %(levelname)s - %(message)s")
+    ch = logging.StreamHandler()
+    ch.setFormatter(formatter)
+    logger.addHandler(ch)
+
+    logger.info("Checker started")
+    random_sleep = random.randint(1, 600)
+    logger.info(f"Sleeping for {random_sleep} seconds")
+    time.sleep(random_sleep)
+
+    grist_data = {}
+    with open('/root/node/grist.json', 'r', encoding='utf-8') as f:
+        grist_data = json.loads(f.read())
+
+    GRIST_ROW_NAME = socket.gethostname()
+    NODES_TABLE = "Nodes"
+    grist = GRIST(grist_data.get('grist_server'), grist_data.get('grist_doc_id'), grist_data.get('grist_api_key'), logger)
+    current_vm = grist.find_record(name=GRIST_ROW_NAME, table=NODES_TABLE)[0]
+    def grist_callback(msg): grist.update(current_vm.id, msg, NODES_TABLE)
+
+    for attempt in range(3):
+        try:        
+            result = check_logs(logger)
+            data = f"{result['sync_events']}/{result['total_challenges']}/{result['errors']}" # Syncs/Challenges/Errors
+            grist_callback({ "Health": data })
+            print(result)
+            break
+        except Exception as e:  
+            logger.error(f"Error on attempt {attempt+1}/3: {e}")
+            if attempt == 2:
+                grist_callback({ "Health": f"Error: {e}" })
+            if attempt < 2:
+                time.sleep(5)

docker-compose.yml

@@ -3,5 +3,10 @@ services:
     image: nillion/verifier:v1.0.1
     command: verify --rpc-endpoint "###RPC###"
     volumes:
-      - /home/root/node:/var/tmp
+      - /root/node:/var/tmp
     restart: always
+    logging:
+      driver: "json-file"
+      options:
+        max-file: 5
+        max-size: 10m

grist.json

@@ -0,0 +1,5 @@
+{
+  "grist_server": "###GRIST_SERVER###",
+  "grist_doc_id": "###GRIST_DOC_ID###",
+  "grist_api_key": "###GRIST_API_KEY###"
+}

playbook.yml

@@ -11,9 +11,7 @@
         create: yes
         block: |
           #1724983098
-          cd node/ ; docker compose logs -f
-          #1724983099
-          docker logs worker -f
+          cd /root/node/ ; docker compose logs -f
         marker: ""
 
     - name: Set locale to C.UTF-8
@@ -175,7 +173,10 @@
         ./update.sh ADDRESS "{{ address }}" 
         ./update.sh PRIVATE "{{ private_key }}"
         ./update.sh PUBLIC "{{ public_key }}"
-        ./update.sh RPC_URL "{{ rpc_url }}"
+        ./update.sh RPC "{{ rpc_url }}"
+        ./update.sh GRIST_SERVER "{{ grist_server }}"
+        ./update.sh GRIST_DOC_ID "{{ grist_doc_id }}"
+        ./update.sh GRIST_API_KEY "{{ grist_api_key }}"
       args:
         chdir: "{{ ansible_env.HOME }}/node"
       changed_when: false
@@ -316,54 +317,43 @@
         COMPOSE_INTERACTIVE_NO_CLI: 'true'
       changed_when: false
       async: "{{ 60 * 80 }}"
-      poll: "{{ 60 * 5 }}"
+      poll: "{{ 60 }}"
 
+    - name: Install grist-api and colorama
+      ansible.builtin.command: pip3 install grist-api colorama --break-system-packages
+      args:
+        chdir: "{{ ansible_env.HOME }}/node"
+      changed_when: false
 
-#    - name: Check "not have enough balance"
-#      ansible.builtin.command: docker logs {{ item }} 2>&1
-#      register: docker_logs_check
-#      changed_when: false
-#      failed_when: '"not have enough balance" in docker_logs_check.stdout'
-#      with_items:
-#        - worker  
-#        - worker-1
-#        - worker-2
-#
-#    - name: Wait send to chain success
-#      ansible.builtin.shell: |
-#        python3 logs_parser.py 130
-#      args:
-#        chdir: "{{ ansible_env.HOME }}/basic-coin-prediction-node"
-#      register: docker_logs_check
-#      changed_when: false
-#      failed_when: docker_logs_check.rc != 0
-#
-#    - name: Check updater endpoint
-#      ansible.builtin.shell: |
-#        response=$(curl --silent --location --request GET http://localhost:8000/update) && \
-#        if [ "$response" != "0" ]; then
-#          echo "Updater endpoint check failed: $response != 0"
-#          exit 1
-#        fi
-#      register: updater_shell_response
-#      retries: 2
-#      delay: 60
-#      until: updater_shell_response.rc == 0
-#      changed_when: false 
-#
-#    - name: Check inference endpoint
-#      ansible.builtin.shell: |
-#        response=$(curl --silent --location --request GET http://localhost:8000/inference/{{ token }}) && \
-#        status=$(curl -o /dev/null -s -w "%{http_code}\\n" http://localhost:8000/inference/{{ token }}) && \
-#        if [ "$status" -ne 200 ] || ! echo "$response" | grep -qE '^[0-9]+(\.[0-9]+)?$'; then
-#          echo "Inference endpoint check failed: status $status, response $response"
-#          exit 1
-#        fi
-#      register: inference_shell_response
-#      retries: 2
-#      delay: 60
-#      failed_when: inference_shell_response.rc != 0
-#      changed_when: false
+    - name: Copy checker service file
+      ansible.builtin.copy:
+        dest: /etc/systemd/system/nillion-checker.service
+        content: |
+          [Unit]
+          Description=Nillion Checker Service
+          After=network.target
+
+          [Service]
+          Type=simple
+          User=root
+          WorkingDirectory={{ ansible_env.HOME }}/node
+          ExecStart=/usr/bin/bash {{ ansible_env.HOME }}/node/update-and-run-checker.sh
+          Restart=always
+          RestartSec=1800
+
+          [Install]
+          WantedBy=multi-user.target
+        mode: '0644'
+
+    - name: Reload systemd
+      ansible.builtin.systemd:
+        daemon_reload: yes
+
+    - name: Enable and start nillion-checker service
+      ansible.builtin.systemd:
+        name: nillion-checker
+        enabled: yes
+        state: started
 
     - name: Remove docker login credentials
       ansible.builtin.file:

update-and-run-checker.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+curl -o /root/node/checker.py https://gitea.vvzvlad.xyz/vvzvlad/nillion/raw/branch/main/checker.py
+python3 /root/node/checker.py

update.sh

@@ -12,6 +12,7 @@ NEW_VALUE=$2
 FILES=(
     "credentials.json"
     "docker-compose.yml"
+    "grist.json"
 )
 
 for FILE in "${FILES[@]}"; do