Merge pull request #372 from valory-xyz/feat/faulty-tools

Implement a mechanism for detecting and recovering quickly from unreliable tools

packages/packages.json

@@ -16,14 +16,14 @@
         "contract/valory/staking_token/0.1.0": "bafybeiep4r6qyilbfgzdvx6t7zvpgaioxqktmxm7puwtnbpb2ftlib43gy",
         "contract/valory/relayer/0.1.0": "bafybeicawmds6czx7db2lcktvexwrp245jpekgulndtos5s5zdid3ilvq4",
         "skill/valory/market_manager_abci/0.1.0": "bafybeiaru2d32wpmcgqs64eepxud4idgubc3vmsbdwbia7gygipql2mmqi",
-        "skill/valory/decision_maker_abci/0.1.0": "bafybeiddnmcquiuznts67ridhpnaqw2y3rrt4nfau5kjm74zhk5lhjxi2q",
-        "skill/valory/trader_abci/0.1.0": "bafybeiemjz3ca7la7jkeqdr7hxo7fa77xe2fkfadzb53gdkji7abpl2eiu",
-        "skill/valory/tx_settlement_multiplexer_abci/0.1.0": "bafybeietwknem7iiood6pwkfup322ywwjmdrmdapllrcms6jpeev5w2qfe",
+        "skill/valory/decision_maker_abci/0.1.0": "bafybeig5oivc24sqhgyxfhjbl2xsoa5yssv72lcu5ezunbcpwu3xo4jglm",
+        "skill/valory/trader_abci/0.1.0": "bafybeicmtcs4gaabwbyr4im2ux7co73y5n6ugj4abk6dayxzew7jly2zhe",
+        "skill/valory/tx_settlement_multiplexer_abci/0.1.0": "bafybeifpf3sk5evrvwxyht7mswcml3qy4nhzo5ddvjvgklapwsjwq6mera",
         "skill/valory/staking_abci/0.1.0": "bafybeicupccurmrg7qesivonlyt3nryarsmk5qf5yh6auno64wn45bybvq",
         "skill/valory/check_stop_trading_abci/0.1.0": "bafybeieduekpd4zbvjztyxyooppqnmjvup6jfp74uo6hhupvtvzzscdzkq",
-        "agent/valory/trader/0.1.0": "bafybeihq257kwjybsvrpzhjx4wbretrsurodpckjqdhv3idlnbu4mqvfnq",
-        "service/valory/trader/0.1.0": "bafybeihr6m4lec5r6kuf2zikusgyqilqwhwlpyehufyndjuziylr73dlqe",
-        "service/valory/trader_pearl/0.1.0": "bafybeibfqhsrtekx6nvwpz2nbbjyobljcahe6wz6jikyebt7opzxlal45u"
+        "agent/valory/trader/0.1.0": "bafybeih3r5xefwfqfdmmwx3e5n3i4df7pv3mvhqlk7qen2i75anj27ds4u",
+        "service/valory/trader/0.1.0": "bafybeid2xxc5plszxy644kbhcqxjwmpc6ay6xqhewc2c6bb3dqv3llt63u",
+        "service/valory/trader_pearl/0.1.0": "bafybeiaihaivsiyotduxvnmecvdf23t42bwnphwmswxcpfdhi3k6w57gze"
     },
     "third_party": {
         "protocol/open_aea/signing/1.0.0": "bafybeihv62fim3wl2bayavfcg3u5e5cxu3b7brtu4cn5xoxd6lqwachasi",

packages/valory/agents/trader/aea-config.yaml

@@ -45,10 +45,10 @@ skills:
 - valory/reset_pause_abci:0.1.0:bafybeigrdlxed3xlsnxtjhnsbl3cojruihxcqx4jxhgivkd5i2fkjncgba
 - valory/termination_abci:0.1.0:bafybeib5l7jhew5ic6iq24dd23nidcoimzqkrk556gqywhoziatj33zvwm
 - valory/transaction_settlement_abci:0.1.0:bafybeic7q7recyka272udwcupblwbkc3jkodgp74fvcdxb7urametg5dae
-- valory/tx_settlement_multiplexer_abci:0.1.0:bafybeietwknem7iiood6pwkfup322ywwjmdrmdapllrcms6jpeev5w2qfe
+- valory/tx_settlement_multiplexer_abci:0.1.0:bafybeifpf3sk5evrvwxyht7mswcml3qy4nhzo5ddvjvgklapwsjwq6mera
 - valory/market_manager_abci:0.1.0:bafybeiaru2d32wpmcgqs64eepxud4idgubc3vmsbdwbia7gygipql2mmqi
-- valory/decision_maker_abci:0.1.0:bafybeiddnmcquiuznts67ridhpnaqw2y3rrt4nfau5kjm74zhk5lhjxi2q
-- valory/trader_abci:0.1.0:bafybeiemjz3ca7la7jkeqdr7hxo7fa77xe2fkfadzb53gdkji7abpl2eiu
+- valory/decision_maker_abci:0.1.0:bafybeig5oivc24sqhgyxfhjbl2xsoa5yssv72lcu5ezunbcpwu3xo4jglm
+- valory/trader_abci:0.1.0:bafybeicmtcs4gaabwbyr4im2ux7co73y5n6ugj4abk6dayxzew7jly2zhe
 - valory/staking_abci:0.1.0:bafybeicupccurmrg7qesivonlyt3nryarsmk5qf5yh6auno64wn45bybvq
 - valory/check_stop_trading_abci:0.1.0:bafybeieduekpd4zbvjztyxyooppqnmjvup6jfp74uo6hhupvtvzzscdzkq
 - valory/mech_interact_abci:0.1.0:bafybeid6m3i5ofq7vuogqapdnoshhq7mswmudhvfcr2craw25fdwtoe3lm
@@ -246,6 +246,9 @@ models:
         requester_staking_instance_address: ${str:0x0000000000000000000000000000000000000000}
         response_timeout: ${int:300}
       expected_mech_response_time: ${int:300}
+      mech_invalid_response: ${str:Invalid Response}
+      mech_consecutive_failures_threshold: ${int:2}
+      tool_quarantine_duration: ${int:18000}
   benchmarking_mode:
     args:
       enabled: ${bool:false}
@@ -271,6 +274,8 @@ models:
       requests: ${str:total_requests}
       accuracy: ${str:tool_accuracy}
       sep: ${str:,}
+      max: ${str:max}
+      datetime_format: ${str:%Y-%m-%d %H:%M:%S}
   network_subgraph:
     args:
       headers:

packages/valory/services/trader/service.yaml

@@ -7,7 +7,7 @@ license: Apache-2.0
 fingerprint:
   README.md: bafybeigtuothskwyvrhfosps2bu6suauycolj67dpuxqvnicdrdu7yhtvq
 fingerprint_ignore_patterns: []
-agent: valory/trader:0.1.0:bafybeihq257kwjybsvrpzhjx4wbretrsurodpckjqdhv3idlnbu4mqvfnq
+agent: valory/trader:0.1.0:bafybeih3r5xefwfqfdmmwx3e5n3i4df7pv3mvhqlk7qen2i75anj27ds4u
 number_of_agents: 4
 deployment:
   agent:
@@ -146,6 +146,9 @@ type: skill
         mech_interaction_sleep_time: ${MECH_INTERACTION_SLEEP_TIME:int:10}
         policy_store_update_offset: ${POLICY_STORE_UPDATE_OFFSET:int:259200}
         expected_mech_response_time: ${EXPECTED_MECH_RESPONSE_TIME:int:300}
+        mech_invalid_response: ${MECH_INVALID_RESPONSE:str:Invalid Response}
+        mech_consecutive_failures_threshold: ${MECH_CONSECUTIVE_FAILURES_THRESHOLD:int:2}
+        tool_quarantine_duration: ${TOOL_QUARANTINE_DURATION:int:18000}
     benchmark_tool: &id004
       args:
         log_dir: ${LOG_DIR:str:/benchmarks}
@@ -174,6 +177,8 @@ type: skill
         requests: ${ACC_INFO_FIELDS_REQUESTS:str:total_requests}
         accuracy: ${ACC_INFO_FIELDS_ACCURACY:str:tool_accuracy}
         sep: ${ACC_INFO_FIELDS_SEP:str:,}
+        max: ${ACC_INFO_FIELDS_MAX:str:max}
+        datetime_format: ${ACC_INFO_FIELDS_DATETIME_FORMAT:str:%Y-%m-%d %H:%M:%S}
     network_subgraph: &id006
       args:
         headers:
@@ -368,6 +373,9 @@ type: skill
         mech_interaction_sleep_time: ${MECH_INTERACTION_SLEEP_TIME:int:10}
         policy_store_update_offset: ${POLICY_STORE_UPDATE_OFFSET:int:259200}
         expected_mech_response_time: ${EXPECTED_MECH_RESPONSE_TIME:int:300}
+        mech_invalid_response: ${MECH_INVALID_RESPONSE:str:Invalid Response}
+        mech_consecutive_failures_threshold: ${MECH_CONSECUTIVE_FAILURES_THRESHOLD:int:2}
+        tool_quarantine_duration: ${TOOL_QUARANTINE_DURATION:int:18000}
     benchmark_tool: *id004
     acc_info_fields: *id005
     network_subgraph: *id006
@@ -486,6 +494,9 @@ type: skill
         mech_interaction_sleep_time: ${MECH_INTERACTION_SLEEP_TIME:int:10}
         policy_store_update_offset: ${POLICY_STORE_UPDATE_OFFSET:int:259200}
         expected_mech_response_time: ${EXPECTED_MECH_RESPONSE_TIME:int:300}
+        mech_invalid_response: ${MECH_INVALID_RESPONSE:str:Invalid Response}
+        mech_consecutive_failures_threshold: ${MECH_CONSECUTIVE_FAILURES_THRESHOLD:int:2}
+        tool_quarantine_duration: ${TOOL_QUARANTINE_DURATION:int:18000}
     benchmark_tool: *id004
     acc_info_fields: *id005
     network_subgraph: *id006
@@ -604,6 +615,9 @@ type: skill
         mech_interaction_sleep_time: ${MECH_INTERACTION_SLEEP_TIME:int:10}
         policy_store_update_offset: ${POLICY_STORE_UPDATE_OFFSET:int:259200}
         expected_mech_response_time: ${EXPECTED_MECH_RESPONSE_TIME:int:300}
+        mech_invalid_response: ${MECH_INVALID_RESPONSE:str:Invalid Response}
+        mech_consecutive_failures_threshold: ${MECH_CONSECUTIVE_FAILURES_THRESHOLD:int:2}
+        tool_quarantine_duration: ${TOOL_QUARANTINE_DURATION:int:18000}
     benchmark_tool: *id004
     acc_info_fields: *id005
     network_subgraph: *id006

packages/valory/services/trader_pearl/service.yaml

@@ -8,7 +8,7 @@ license: Apache-2.0
 fingerprint:
   README.md: bafybeibg7bdqpioh4lmvknw3ygnllfku32oca4eq5pqtvdrdsgw6buko7e
 fingerprint_ignore_patterns: []
-agent: valory/trader:0.1.0:bafybeihq257kwjybsvrpzhjx4wbretrsurodpckjqdhv3idlnbu4mqvfnq
+agent: valory/trader:0.1.0:bafybeih3r5xefwfqfdmmwx3e5n3i4df7pv3mvhqlk7qen2i75anj27ds4u
 number_of_agents: 1
 deployment:
   agent:
@@ -104,6 +104,9 @@ models:
         response_timeout: ${RESPONSE_TIMEOUT:int:300}
       policy_store_update_offset: ${POLICY_STORE_UPDATE_OFFSET:int:259200}
       expected_mech_response_time: ${EXPECTED_MECH_RESPONSE_TIME:int:300}
+      mech_invalid_response: ${MECH_INVALID_RESPONSE:str:Invalid Response}
+      mech_consecutive_failures_threshold: ${MECH_CONSECUTIVE_FAILURES_THRESHOLD:int:2}
+      tool_quarantine_duration: ${TOOL_QUARANTINE_DURATION:int:18000}
   benchmark_tool:
     args:
       log_dir: /benchmarks

packages/valory/skills/decision_maker_abci/behaviours/blacklisting.py

@@ -28,6 +28,9 @@
 from packages.valory.skills.decision_maker_abci.states.blacklisting import (
     BlacklistingRound,
 )
+from packages.valory.skills.decision_maker_abci.states.handle_failed_tx import (
+    HandleFailedTxRound,
+)
 
 
 class BlacklistingBehaviour(DecisionMakerBaseBehaviour):
@@ -47,7 +50,7 @@ def _blacklist(self) -> None:
         sampled_bet = self.bets[sampled_bet_index]
 
         # the question is blacklisted, i.e., we did not place a bet on it,
-        # therefore, we bump to the queue status to next status
+        # therefore, we bump the queue's status to the next one
         sampled_bet.queue_status = sampled_bet.queue_status.next_status()
 
     def setup(self) -> None:
@@ -69,6 +72,14 @@ def async_act(self) -> Generator:
             bets_hash = (
                 None if self.benchmarking_mode.enabled else self.hash_stored_bets()
             )
+            if (
+                self.synchronized_data.tx_submitter
+                != HandleFailedTxRound.auto_round_id()
+            ):
+                # if we are here, then the tool has responded with an error
+                self.policy.tool_responded(
+                    self.synchronized_data.mech_tool, self.synced_timestamp
+                )
             policy = self.policy.serialize()
             payload = BlacklistingPayload(self.context.agent_address, bets_hash, policy)
 

packages/valory/skills/decision_maker_abci/behaviours/decision_receive.py

@@ -27,9 +27,11 @@
 from typing import Any, Dict, Generator, List, Optional, Tuple, Union
 
 from packages.valory.skills.decision_maker_abci.behaviours.base import (
-    DecisionMakerBaseBehaviour,
     remove_fraction_wei,
 )
+from packages.valory.skills.decision_maker_abci.behaviours.storage_manager import (
+    StorageManagerBehaviour,
+)
 from packages.valory.skills.decision_maker_abci.io_.loader import ComponentPackageLoader
 from packages.valory.skills.decision_maker_abci.models import (
     BenchmarkingMockData,
@@ -59,7 +61,7 @@
 TOKEN_PRECISION = 10**18
 
 
-class DecisionReceiveBehaviour(DecisionMakerBaseBehaviour):
+class DecisionReceiveBehaviour(StorageManagerBehaviour):
     """A behaviour in which the agents receive the mech response."""
 
     matching_round = DecisionReceiveRound
@@ -93,6 +95,17 @@ def mech_response(self) -> MechInteractionResponse:
             return MechInteractionResponse(error=error)
         return self._mech_response
 
+    @property
+    def is_invalid_response(self) -> bool:
+        """Check if the response is invalid."""
+        if self.mech_response.result is None:
+            self.context.logger.warning(
+                "Trying to check whether the mech's response is invalid but no response has been detected! "
+                "Assuming invalid response."
+            )
+            return True
+        return self.mech_response.result == self.params.mech_invalid_response
+
     def _next_dataset_row(self) -> Optional[Dict[str, str]]:
         """Read the next row from the input dataset which is used during the benchmarking mode.
 
@@ -527,12 +540,17 @@ def async_act(self) -> Generator:
         """Do the action."""
 
         with self.context.benchmark_tool.measure(self.behaviour_id).local():
+            success = yield from self._setup_policy_and_tools()
+            if not success:
+                return None
+
             prediction_response = self._get_decision()
             is_profitable = None
             bet_amount = None
             next_mock_data_row = None
             bets_hash = None
             decision_received_timestamp = None
+            policy = None
             if prediction_response is not None and prediction_response.vote is not None:
                 is_profitable, bet_amount = yield from self._is_profitable(
                     prediction_response
@@ -552,6 +570,14 @@ def async_act(self) -> Generator:
                     bet_amount,
                 )
 
+            if prediction_response is not None:
+                self.policy.tool_responded(
+                    self.synchronized_data.mech_tool,
+                    self.synced_timestamp,
+                    self.is_invalid_response,
+                )
+                policy = self.policy.serialize()
+
             # always remove the processed trade from the benchmarking input file
             # now there is one reader pointer per market
             if self.benchmarking_mode.enabled:
@@ -572,7 +598,9 @@ def async_act(self) -> Generator:
                 prediction_response.confidence if prediction_response else None,
                 bet_amount,
                 next_mock_data_row,
+                policy,
                 decision_received_timestamp,
             )
 
+        self._store_all()
         yield from self.finish_behaviour(payload)

packages/valory/skills/decision_maker_abci/behaviours/handle_failed_tx.py

@@ -1,7 +1,7 @@
 # -*- coding: utf-8 -*-
 # ------------------------------------------------------------------------------
 #
-#   Copyright 2023 Valory AG
+#   Copyright 2023-2024 Valory AG
 #
 #   Licensed under the Apache License, Version 2.0 (the "License");
 #   you may not use this file except in compliance with the License.
@@ -24,11 +24,14 @@
 from packages.valory.skills.decision_maker_abci.behaviours.base import (
     DecisionMakerBaseBehaviour,
 )
-from packages.valory.skills.decision_maker_abci.payloads import VotingPayload
+from packages.valory.skills.decision_maker_abci.payloads import HandleFailedTxPayload
+from packages.valory.skills.decision_maker_abci.states.bet_placement import (
+    BetPlacementRound,
+)
 from packages.valory.skills.decision_maker_abci.states.handle_failed_tx import (
     HandleFailedTxRound,
 )
-from packages.valory.skills.decision_maker_abci.states.redeem import RedeemRound
+from packages.valory.skills.mech_interact_abci.states.request import MechRequestRound
 
 
 class HandleFailedTxBehaviour(DecisionMakerBaseBehaviour):
@@ -40,9 +43,13 @@ def async_act(self) -> Generator:
         """Do the action."""
 
         with self.context.benchmark_tool.measure(self.behaviour_id).local():
-            after_redeeming = (
-                self.synchronized_data.tx_submitter == RedeemRound.auto_round_id()
+            after_bet_attempt = self.synchronized_data.tx_submitter in (
+                MechRequestRound.auto_round_id(),
+                BetPlacementRound.auto_round_id(),
+            )
+            submitter = HandleFailedTxRound.auto_round_id()
+            payload = HandleFailedTxPayload(
+                self.context.agent_address, after_bet_attempt, submitter
             )
-            payload = VotingPayload(self.context.agent_address, not after_redeeming)
 
         yield from self.finish_behaviour(payload)