Error minutes

Find the minutes in which an API gateway was struggling, from its request log.

Each log entry is a string "ts id path status" with single spaces between the four fields: ts is the time of the request in whole seconds since midnight, id identifies the request, path is the requested path and status is the HTTP status code that was returned. Entries are not necessarily in time order.

Build the picture of each minute like this, in this order:

1. Ignore every entry whose path starts with /health.
2. A client may retry a request, which creates several entries with the same id. Only the last attempt counts, the entry with the greatest ts (if two attempts share that ts, the one that appears later in the list). The earlier attempts are ignored entirely, even if they fall in a different minute.
3. Drop the remaining entries with status 429. They are rejections by the rate limiter and say nothing about the service.
4. Each remaining entry belongs to minute ts // 60 (minute 0 covers seconds 0 to 59, minute 1 covers 60 to 119, and so on). It is an error if its status is 500 or higher.

A minute is degraded if it has at least 5 remaining entries and strictly more than 20% of them are errors.

Input: entries, a list of strings in the format above.

Output: the degraded minutes in ascending order, each written as its start time "HH:MM" (24-hour, zero padded), as a list of strings. If no minute is degraded, return [].