Add remaining health checks and shutdown on critical errors

apps/cic-eth/cic_eth/admin/ctrl.py

@@ -2,7 +2,7 @@
 import datetime
 import logging
 
-# third-party imports
+# external imports
 import celery
 from chainlib.eth.constant import ZERO_ADDRESS
 from chainlib.chain import ChainSpec
@@ -145,3 +145,9 @@ def check_lock(chained_input, chain_spec_dict, lock_flags, address=None):
     session.flush()
     session.close()
     return chained_input
+
+
+@celery_app.task()
+def shutdown(message):
+    logg.critical('shutdown called: {}'.format(message))
+    celery_app.control.shutdown() #broadcast('shutdown')

apps/cic-eth/cic_eth/admin/token.py

@@ -0,0 +1,19 @@
+# standard imports
+import logging
+
+# external imports
+import celery
+
+# local imports
+from cic_eth.task import BaseTask
+
+celery_app = celery.current_app
+logg = logging.getLogger()
+
+
+@celery_app.task(bind=True, base=BaseTask)
+def default_token(self):
+    return {
+            'symbol': self.default_token_symbol,
+            'address': self.default_token_address,
+            }

apps/cic-eth/cic_eth/check/redis.py

@@ -0,0 +1,18 @@
+# external imports
+import redis
+import os
+
+
+def health(*args, **kwargs):
+    r = redis.Redis(
+            host=kwargs['config'].get('REDIS_HOST'),
+            port=kwargs['config'].get('REDIS_PORT'),
+            db=kwargs['config'].get('REDIS_DB'),
+            )
+    try:
+        r.set(kwargs['unit'], os.getpid())
+    except redis.connection.ConnectionError:
+        return False
+    except redis.connection.ResponseError:
+        return False
+    return True

apps/cic-eth/cic_eth/error.py

@@ -48,6 +48,8 @@ class RoleMissingError(Exception):
     pass
 
 
+
+
 class IntegrityError(Exception):
     """Exception raised to signal irregularities with deduplication and ordering of tasks
 
@@ -62,15 +64,19 @@ class LockedError(Exception):
     pass
 
 
-class SignerError(Exception):
+class SeppukuError(Exception):
+    """Exception base class for all errors that should cause system shutdown
+
+    """
+
+
+class SignerError(SeppukuError):
     """Exception raised when signer is unavailable or generates an error
 
     """
     pass
 
 
-class EthError(Exception):
+class RoleAgencyError(SeppukuError):
-    """Exception raised when unspecified error from evm node is encountered
+    """Exception raise when a role cannot perform its function. This is a critical exception
-
     """
-    pass

apps/cic-eth/cic_eth/eth/account.py

@@ -4,10 +4,10 @@ import logging
 # external imports
 import celery
 from erc20_single_shot_faucet import SingleShotFaucet as Faucet
-from chainlib.eth.constant import ZERO_ADDRESS
 from hexathon import (
         strip_0x,
         )
+from chainlib.eth.constant import ZERO_ADDRESS
 from chainlib.connection import RPCConnection
 from chainlib.eth.sign import (
         new_account,
@@ -19,6 +19,7 @@ from chainlib.eth.tx import (
         unpack,
         )
 from chainlib.chain import ChainSpec
+from chainlib.error import JSONRPCException
 from eth_accounts_index import AccountRegistry
 from sarafu_faucet import MinterFaucet as Faucet
 from chainqueue.db.models.tx import TxCache
@@ -70,11 +71,18 @@ def create(self, password, chain_spec_dict):
     a = None
     conn = RPCConnection.connect(chain_spec, 'signer')
     o = new_account()
-    a = conn.do(o)
+    try:
+        a = conn.do(o)
+    except ConnectionError as e:
+        raise SignerError(e)
+    except FileNotFoundError as e:
+        raise SignerError(e)
     conn.disconnect()
 
+    # TODO: It seems infeasible that a can be None in any case, verify
     if a == None:
         raise SignerError('create account')
+        
     logg.debug('created account {}'.format(a))
 
     # Initialize nonce provider record for account
@@ -219,21 +227,22 @@ def have(self, account, chain_spec_dict):
     """
     chain_spec = ChainSpec.from_dict(chain_spec_dict)
     o = sign_message(account, '0x2a')
-    try:
+    conn = RPCConnection.connect(chain_spec, 'signer')
-        conn = RPCConnection.connect(chain_spec, 'signer')
-    except Exception as e:
-        logg.debug('cannot sign with {}: {}'.format(account, e))
-        return None
 
     try:
         conn.do(o)
-        conn.disconnect()
+    except ConnectionError as e:
-        return account
+        raise SignerError(e)
-    except Exception as e:
+    except FileNotFoundError as e:
+        raise SignerError(e)
+    except JSONRPCException as e:
         logg.debug('cannot sign with {}: {}'.format(account, e))
         conn.disconnect()
         return None
    
+    conn.disconnect()
+    return account
+
 
 @celery_app.task(bind=True, base=CriticalSQLAlchemyTask)
 def set_role(self, tag, address, chain_spec_dict):

apps/cic-eth/cic_eth/eth/erc20.py

@@ -108,7 +108,13 @@ def transfer(self, tokens, holder_address, receiver_address, value, chain_spec_d
     nonce_oracle = CustodialTaskNonceOracle(holder_address, self.request.root_id, session=session)
     gas_oracle = self.create_gas_oracle(rpc, MaxGasOracle.gas)
     c = ERC20(chain_spec, signer=rpc_signer, gas_oracle=gas_oracle, nonce_oracle=nonce_oracle)
-    (tx_hash_hex, tx_signed_raw_hex) = c.transfer(t['address'], holder_address, receiver_address, value, tx_format=TxFormat.RLP_SIGNED)
+    try:
+        (tx_hash_hex, tx_signed_raw_hex) = c.transfer(t['address'], holder_address, receiver_address, value, tx_format=TxFormat.RLP_SIGNED)
+    except FileNotFoundError as e:
+        raise SignerError(e)
+    except ConnectionError as e:
+        raise SignerError(e)
+    
 
     rpc_signer.disconnect()
     rpc.disconnect()
@@ -171,7 +177,12 @@ def approve(self, tokens, holder_address, spender_address, value, chain_spec_dic
     nonce_oracle = CustodialTaskNonceOracle(holder_address, self.request.root_id, session=session)
     gas_oracle = self.create_gas_oracle(rpc, MaxGasOracle.gas)
     c = ERC20(chain_spec, signer=rpc_signer, gas_oracle=gas_oracle, nonce_oracle=nonce_oracle)
-    (tx_hash_hex, tx_signed_raw_hex) = c.approve(t['address'], holder_address, spender_address, value, tx_format=TxFormat.RLP_SIGNED)
+    try:
+        (tx_hash_hex, tx_signed_raw_hex) = c.approve(t['address'], holder_address, spender_address, value, tx_format=TxFormat.RLP_SIGNED)
+    except FileNotFoundError as e:
+        raise SignerError(e)
+    except ConnectionError as e:
+        raise SignerError(e)
 
     rpc_signer.disconnect()
     rpc.disconnect()

apps/cic-eth/cic_eth/eth/gas.py

@@ -328,7 +328,12 @@ def refill_gas(self, recipient_address, chain_spec_dict):
 
     # build and add transaction
     logg.debug('tx send gas amount {} from provider {} to {}'.format(refill_amount, gas_provider, recipient_address))
-    (tx_hash_hex, tx_signed_raw_hex) = c.create(gas_provider, recipient_address, refill_amount, tx_format=TxFormat.RLP_SIGNED)
+    try:
+        (tx_hash_hex, tx_signed_raw_hex) = c.create(gas_provider, recipient_address, refill_amount, tx_format=TxFormat.RLP_SIGNED)
+    except ConnectionError as e:
+        raise SignerError(e)
+    except FileNotFoundError as e:
+        raise SignerError(e)
     logg.debug('adding queue refill gas tx {}'.format(tx_hash_hex))
     cache_task = 'cic_eth.eth.gas.cache_gas_data'
     register_tx(tx_hash_hex, tx_signed_raw_hex, chain_spec, queue, cache_task=cache_task, session=session)
@@ -404,7 +409,12 @@ def resend_with_higher_gas(self, txold_hash_hex, chain_spec_dict, gas=None, defa
     c = TxFactory(chain_spec, signer=rpc_signer, gas_oracle=gas_oracle)
     logg.debug('change gas price from old {} to new {} for tx {}'.format(tx['gasPrice'], new_gas_price, tx))
     tx['gasPrice'] = new_gas_price
-    (tx_hash_hex, tx_signed_raw_hex) = c.build_raw(tx)
+    try:
+        (tx_hash_hex, tx_signed_raw_hex) = c.build_raw(tx)
+    except ConnectionError as e:
+        raise SignerError(e)
+    except FileNotFoundError as e:
+        raise SignerError(e)
     queue_create(
         chain_spec,
         tx['nonce'],

apps/cic-eth/cic_eth/runnable/daemons/tasker.py

@@ -11,8 +11,14 @@ import websocket
 # external imports
 import celery
 import confini
-from chainlib.connection import RPCConnection
+from chainlib.connection import (
-from chainlib.eth.connection import EthUnixSignerConnection
+        RPCConnection,
+        ConnType,
+        )
+from chainlib.eth.connection import (
+        EthUnixSignerConnection,
+        EthHTTPSignerConnection,
+        )
 from chainlib.chain import ChainSpec
 from chainqueue.db.models.otx import Otx
 from cic_eth_registry.error import UnknownContractError
@@ -143,8 +149,10 @@ else:
         })
 
 chain_spec = ChainSpec.from_chain_str(config.get('CIC_CHAIN_SPEC'))
+RPCConnection.register_constructor(ConnType.UNIX, EthUnixSignerConnection, 'signer')
+RPCConnection.register_constructor(ConnType.HTTP, EthHTTPSignerConnection, 'signer')
+RPCConnection.register_constructor(ConnType.HTTP_SSL, EthHTTPSignerConnection, 'signer')
 RPCConnection.register_location(config.get('ETH_PROVIDER'), chain_spec, 'default')
-#RPCConnection.register_location(config.get('SIGNER_SOCKET_PATH'), chain_spec, 'signer', constructor=EthUnixSignerConnection)
 RPCConnection.register_location(config.get('SIGNER_SOCKET_PATH'), chain_spec, 'signer')
 
 Otx.tracing = config.true('TASKS_TRACE_QUEUE_STATUS')
@@ -152,7 +160,7 @@ Otx.tracing = config.true('TASKS_TRACE_QUEUE_STATUS')
 #import cic_eth.checks.gas
 #if not cic_eth.checks.gas.health(config=config):
 #    raise RuntimeError()
-liveness.linux.load(health_modules, rundir=config.get('CIC_RUN_DIR'), config=config)
+liveness.linux.load(health_modules, rundir=config.get('CIC_RUN_DIR'), config=config, unit='cic-eth-tasker')
 
 def main():
     argv = ['worker']
@@ -195,11 +203,12 @@ def main():
 
     BaseTask.default_token_symbol = config.get('CIC_DEFAULT_TOKEN_SYMBOL')
     BaseTask.default_token_address = registry.by_name(BaseTask.default_token_symbol)
+    BaseTask.run_dir = config.get('CIC_RUN_DIR')
     logg.info('default token set to {} {}'.format(BaseTask.default_token_symbol, BaseTask.default_token_address))
    
-    liveness.linux.set()
+    liveness.linux.set(rundir=config.get('CIC_RUN_DIR'))
     current_app.worker_main(argv)
-    liveness.linux.reset()
+    liveness.linux.reset(rundir=config.get('CIC_RUN_DIR'))
 
 
 @celery.signals.eventlet_pool_postshutdown.connect

apps/cic-eth/cic_eth/runnable/info.py

@@ -0,0 +1,65 @@
+#!python3
+
+# SPDX-License-Identifier: GPL-3.0-or-later
+
+# standard imports
+import logging
+import argparse
+import os
+
+# external imports
+import confini
+import celery
+
+# local imports
+from cic_eth.api import Api
+
+logging.basicConfig(level=logging.WARNING)
+logg = logging.getLogger()
+
+default_format = 'terminal'
+default_config_dir = os.environ.get('CONFINI_DIR', '/usr/local/etc/cic')
+
+
+argparser = argparse.ArgumentParser()
+argparser.add_argument('-i', '--chain-spec', dest='i', type=str, help='chain spec')
+argparser.add_argument('-c', type=str, default=default_config_dir, help='config root to use')
+argparser.add_argument('-q', type=str, default='cic-eth', help='celery queue to submit transaction tasks to')
+argparser.add_argument('--env-prefix', default=os.environ.get('CONFINI_ENV_PREFIX'), dest='env_prefix', type=str, help='environment prefix for variables to overwrite configuration')
+argparser.add_argument('-v', action='store_true', help='Be verbose')
+argparser.add_argument('-vv', help='be more verbose', action='store_true')
+args = argparser.parse_args()
+
+if args.v == True:
+    logging.getLogger().setLevel(logging.INFO)
+elif args.vv == True:
+    logging.getLogger().setLevel(logging.DEBUG)
+
+config_dir = os.path.join(args.c)
+os.makedirs(config_dir, 0o777, True)
+config = confini.Config(config_dir, args.env_prefix)
+config.process()
+args_override = {
+        'CIC_CHAIN_SPEC': getattr(args, 'i'),
+        }
+config.dict_override(args_override, 'cli args')
+config.censor('PASSWORD', 'DATABASE')
+config.censor('PASSWORD', 'SSL')
+logg.debug('config loaded from {}:\n{}'.format(config_dir, config))
+
+
+celery_app = celery.Celery(broker=config.get('CELERY_BROKER_URL'), backend=config.get('CELERY_RESULT_URL'))
+
+queue = args.q
+
+api = Api(config.get('CIC_CHAIN_SPEC'), queue=queue)
+
+def main():
+    t = api.default_token()
+    token_info = t.get()
+    print('Default token symbol: {}'.format(token_info['symbol']))
+    print('Default token address: {}'.format(token_info['address']))
+
+
+if __name__ == '__main__':
+    main()

apps/cic-eth/cic_eth/task.py

@@ -10,15 +10,13 @@ import sqlalchemy
 from chainlib.eth.constant import ZERO_ADDRESS
 from chainlib.eth.nonce import RPCNonceOracle
 from chainlib.eth.gas import RPCGasOracle
+import liveness.linux
 
 # local imports
-from cic_eth.error import (
+from cic_eth.error import SeppukuError
-        SignerError,
-        EthError,
-        )
 from cic_eth.db.models.base import SessionBase
 
-logg = logging.getLogger(__name__)
+logg = logging.getLogger().getChild(__name__)
 
 celery_app = celery.current_app
 
@@ -31,6 +29,7 @@ class BaseTask(celery.Task):
     create_gas_oracle = RPCGasOracle
     default_token_address = None
     default_token_symbol = None
+    run_dir = '/run'
 
     def create_session(self):
         return BaseTask.session_func()
@@ -40,6 +39,19 @@ class BaseTask(celery.Task):
         logg.debug('task {} root uuid {}'.format(self.__class__.__name__, self.request.root_id))
         return
 
+
+    def on_failure(self, exc, task_id, args, kwargs, einfo):
+        if isinstance(exc, SeppukuError):
+            liveness.linux.reset(rundir=self.run_dir)
+            logg.critical(einfo)
+            msg = 'received critical exception {}, calling shutdown'.format(str(exc))
+            s = celery.signature(
+                'cic_eth.admin.ctrl.shutdown',
+                [msg],
+                queue=self.request.delivery_info.get('routing_key'),
+                    )
+            s.apply_async()
+
     
 class CriticalTask(BaseTask):
     retry_jitter = True
@@ -69,7 +81,6 @@ class CriticalSQLAlchemyAndWeb3Task(CriticalTask):
         sqlalchemy.exc.TimeoutError,
         requests.exceptions.ConnectionError,
         sqlalchemy.exc.ResourceClosedError,
-        EthError,
         )
     safe_gas_threshold_amount = 2000000000 * 60000 * 3
     safe_gas_refill_amount = safe_gas_threshold_amount * 5 
@@ -80,13 +91,11 @@ class CriticalSQLAlchemyAndSignerTask(CriticalTask):
         sqlalchemy.exc.DatabaseError,
         sqlalchemy.exc.TimeoutError,
         sqlalchemy.exc.ResourceClosedError,
-        SignerError,
         ) 
 
 class CriticalWeb3AndSignerTask(CriticalTask):
     autoretry_for = (
         requests.exceptions.ConnectionError,
-        SignerError,
         )
     safe_gas_threshold_amount = 2000000000 * 60000 * 3
     safe_gas_refill_amount = safe_gas_threshold_amount * 5 
@@ -100,4 +109,4 @@ def hello(self):
 
 @celery_app.task()
 def check_health(self):
-    celery.app.control.shutdown()
+    pass

apps/cic-eth/config/cic.ini

@@ -4,5 +4,5 @@ chain_spec = evm:bloxberg:8996
 tx_retry_delay = 
 trust_address =
 default_token_symbol = GFT
-health_modules = cic_eth.check.db,cic_eth.check.signer,cic_eth.check.gas
+health_modules = cic_eth.check.db,cic_eth.check.redis,cic_eth.check.signer,cic_eth.check.gas
 run_dir = /run

apps/cic-eth/config/docker/cic.ini

@@ -4,5 +4,5 @@ chain_spec = evm:bloxberg:8996
 trust_address = 0xEb3907eCad74a0013c259D5874AE7f22DcBcC95C
 tx_retry_delay = 20
 default_token_symbol = GFT
-health_modules = cic_eth.check.db,cic_eth.check.signer,cic_eth.check.gas
+health_modules = cic_eth.check.db,cic_eth.check.redis,cic_eth.check.signer,cic_eth.check.gas
 run_dir = /run

apps/cic-eth/config/docker/eth.ini

@@ -1,4 +1,3 @@
 [eth]
 provider = http://localhost:63545
-health_modules = cic_eth.check.db,cic_eth.check.gas
 gas_gifter_minimum_balance = 10000000000000000000000

apps/cic-eth/config/eth.ini

@@ -1,4 +1,3 @@
 [eth]
 provider = http://localhost:8545
 gas_gifter_minimum_balance = 10000000000000000000000
-health_modules = cic_eth.check.db,cic_eth.check.gas

apps/cic-eth/requirements.txt

@@ -1,4 +1,4 @@
-cic-base==0.1.2b1
+cic-base==0.1.2b2
 celery==4.4.7
 crypto-dev-signer~=0.4.14b3
 confini~=0.3.6rc3

apps/cic-eth/tests/conftest.py

@@ -3,8 +3,12 @@ import os
 import sys
 import logging
 
+# external imports
+from chainlib.eth.erc20 import ERC20
+
 # local imports
 from cic_eth.api import Api
+from cic_eth.task import BaseTask
 
 script_dir = os.path.dirname(os.path.realpath(__file__))
 root_dir = os.path.dirname(script_dir)
@@ -28,3 +32,26 @@ def api(
     ):
     chain_str = str(default_chain_spec)
     return Api(chain_str, queue=None, callback_param='foo')
+
+
+@pytest.fixture(scope='function')
+def foo_token_symbol(
+    default_chain_spec,
+    foo_token,
+    eth_rpc,
+    contract_roles,
+    ):
+    
+    c = ERC20(default_chain_spec)
+    o = c.symbol(foo_token, sender_address=contract_roles['CONTRACT_DEPLOYER'])
+    r = eth_rpc.do(o)
+    return c.parse_symbol(r)
+
+
+@pytest.fixture(scope='function')
+def default_token(
+        foo_token,
+        foo_token_symbol,
+        ):
+    BaseTask.default_token_symbol = foo_token_symbol
+    BaseTask.default_token_address = foo_token   

apps/cic-eth/tests/fixtures_celery.py

@@ -34,6 +34,7 @@ def celery_includes():
         'cic_eth.admin.ctrl',
         'cic_eth.admin.nonce',
         'cic_eth.admin.debug',
+        'cic_eth.admin.token',
         'cic_eth.eth.account',
         'cic_eth.callbacks.noop',
         'cic_eth.callbacks.http',

apps/cic-eth/tests/unit/admin/test_default_token.py

@@ -0,0 +1,21 @@
+# external imports
+import celery
+
+
+def test_default_token(
+        default_token,
+        celery_session_worker,
+        foo_token,
+        foo_token_symbol,
+        ):
+      
+    s = celery.signature(
+            'cic_eth.admin.token.default_token',
+            [],
+            queue=None,
+            )
+    t = s.apply_async()
+    r = t.get()
+
+    assert r['address'] == foo_token
+    assert r['symbol'] == foo_token_symbol

apps/util/liveness/liveness/linux.py

@@ -42,6 +42,7 @@ def load(check_strs, namespace=default_namespace, rundir='/run', *args, **kwargs
 
 
 def set(error=0, namespace=default_namespace, rundir='/run'):
+    logg.info('liveness SET error {} for namespace {}'.format(error, namespace))
     app_rundir = os.path.join(rundir, namespace)
     f = open(os.path.join(app_rundir, 'error'), 'w')
     f.write(str(error))
@@ -49,6 +50,13 @@ def set(error=0, namespace=default_namespace, rundir='/run'):
 
 
 def reset(namespace=default_namespace, rundir='/run'):
+    logg.info('liveness RESET for namespace {}'.format(namespace))
     app_rundir = os.path.join(rundir, namespace)
-    os.unlink(os.path.join(app_rundir, 'pid'))
+    try:
-    os.unlink(os.path.join(app_rundir, 'error'))
+        os.unlink(os.path.join(app_rundir, 'pid'))
+    except FileNotFoundError:
+        pass
+    try:
+        os.unlink(os.path.join(app_rundir, 'error'))
+    except FileNotFoundError:
+        pass

docker-compose.yml

@@ -240,6 +240,8 @@ services:
       DATABASE_DRIVER: ${DATABASE_DRIVER:-psycopg2}
       DATABASE_DEBUG: ${DATABASE_DEBUG:-0}
       DATABASE_POOL_SIZE: 0
+      REDIS_PORT: 6379
+      REDIS_HOST: redis
       PGPASSWORD: ${DATABASE_PASSWORD:-tralala}
       CIC_CHAIN_SPEC: ${CIC_CHAIN_SPEC:-evm:bloxberg:8996}
       BANCOR_DIR: ${BANCOR_DIR:-/usr/local/share/cic/bancor}