Failed graphql query reported as successful

Our indexer was returning errors for some networks:

http post http://host1/subgraphs/id/QmQohvyxLZLpwouzPF61kTzmLjgawmt6qDWZrPndz1M9EQ Host:graph-mainnet.ellipfra.net Authorization:'XXXXX' query='query MyQuery { _meta { block { number } } }'

HTTP/1.1 200 OK { "graphQLResponse": "{\"errors\":[{\"message\":\"Store error: database unavailable\"}]}" }

The graph-service logs are not hinting of any problem:
{"level":30,"time":1679489726789,"pid":1,"hostname":"indexer-service-6fc7bb878c-ljnmg","name":"IndexerService","indexer":"0x62A0BD1d110FF4E5b793119e95Fc07C9d1Fc8c4a","operator":"0x4ecB19A2aC49C5DecFa5E65B6669C7e7fab5da9D","indexer":"0x62A0BD1d110FF4E5b793119e95Fc07C9d1Fc8c4a","operator":"0x4ecB19A2aC49C5DecFa5E65B6669C7e7fab5da9D","component":"Server","deployment":{"bytes32":"0x24a5b2e65c85a2debf7d9dd783f9c0bc1df52620039e94a2d60cee560834f969","ipfsHash":"QmQohvyxLZLpwouzPF61kTzmLjgawmt6qDWZrPndz1M9EQ"},"msg":"Received free query"} {"level":20,"time":1679489726795,"pid":1,"hostname":"indexer-service-6fc7bb878c-ljnmg","name":"IndexerService","indexer":"0x62A0BD1d110FF4E5b793119e95Fc07C9d1Fc8c4a","operator":"0x4ecB19A2aC49C5DecFa5E65B6669C7e7fab5da9D","indexer":"0x62A0BD1d110FF4E5b793119e95Fc07C9d1Fc8c4a","operator":"0x4ecB19A2aC49C5DecFa5E65B6669C7e7fab5da9D","component":"Server","msg":"POST /subgraphs/id/QmQohvyxLZLpwouzPF61kTzmLjgawmt6qDWZrPndz1M9EQ 200 88 - 6.101 ms"}

The prometheus endpoint is reporting that the query is successful.

graph-indexer v0.20.11

This lack of error reporting is preventing indexers from detecting production issues, as in this situation the problem was only apparent from the gateways.

Start at the graph-service handling of POST /subgraphs/id/... and the Prometheus query-success reporting described in the issue. Reproduce the response containing a GraphQL errors array, then trace how the HTTP result and metric are classified; done means this failure is reported as unsuccessful and the behavior is covered by a regression test.