fix: call await body.text() to avoid undici SocketError: other side closed error

[digitalsadhu]

This PR solves the following error that occurs intermittently (but frequently).

[14:09:16.152] [FATAL] - global uncaught exception - terminating with error. - {
  value: 'SocketError: other side closed\n' +
    '    at Socket.onSocketEnd (/Users/richard.walker@finn.no/src/frontpage-podium/node_modules/undici/lib/client.js:1129:22)\n' +
    '    at Socket.emit (node:events:526:35)\n' +
    '    at endReadableNT (node:internal/streams/readable:1589:12)\n' +
    '    at process.processTicksAndRejections (node:internal/process/task_queues:82:21)'
} - SocketError: other side closed
    at Socket.onSocketEnd (/Users/richard.walker@finn.no/src/frontpage-podium/node_modules/undici/lib/client.js:1129:22)
    at Socket.emit (node:events:526:35)
    at endReadableNT (node:internal/streams/readable:1589:12)
    at process.processTicksAndRejections (node:internal/process/task_queues:82:21)
[14:09:16.158] [ERROR] - Exiting with status 1 in 2 seconds.

There is a massive Undici issue thread thats related nodejs/undici#583 with an unsatisfactory result that concludes that you need to call await body.text(); to avoid the issue which is what this PR does at the expense of essentially buffering up the fetch body before chunking it up again and streaming it as usual.

This is one (sub optimal) way to fix the issue but it does seem reliable. Perhaps there are better ways to fix it? @trygve-lie

[wkillerud]

😿

[leftieFriele]

should we add tests to verify it works? it might be useful if we want to apply a different fix later.

[digitalsadhu]

I think we’ll see what @trygve-lie has to say. This fix is pretty non optimal as it buffers up the stream before chunking it up again which defeats the purpose of the streaming in the first place. More tests is good though of course. If on can be written for an intermittent bug like this.

[trygve-lie]

See: #338.

The real problem seems to be that this only errors when a Podlet responds with a http error. Iow; the way we consume Podlets when they are healthy works as intended but when an Podlet errors we are not consuming the body which Undici require us to do.

This PR works because it consumes the body before we check for http errors but it have the side effect that it also degrades how we consume healthy Podlets.

#388 is a more fine grained in consuming the body.

[trygve-lie]

Fixed in #338.