We send logs from a variety of services running in a Kubernetes cluster to Elasticsearch via Filebeat. Some of these services we develop ourselves, others are third-party. We use dynamic mapping in our indices. We've hit an issue where sometimes a field used by logs from one service happens to share the same name with logs from a difference service, and the type of data in that field is different. For example, in logs from one service, the url field might be a string, but in another, it might be a structured object. We then get errors ingesting the logs, saying:
{
"type": "mapper_parsing_exception",
"reason": "object mapping for [url] tried to parse field [url] as object, but found a concrete value"
}
What strategies might we use to get around these collisions?