Leaked socket in cache/internal/cacheHttpClient

https://github.com/actions/cache/pull/1217 attempted to work around a dangling socket, which became especially apparent when that action migrated to Node 20, as Node v20 started keeping connection alive by default.

This solution worked for actions/cache in the meantime, but caused downstream consumers of to implement the same logic to avoid hanging processes and generated similar issues, such as:

• https://github.com/actions/toolkit/issues/1643
• https://github.com/masci/datadog/pull/33

The leaked socket originates in cache/src/internal/cacheHttpClient:

  const uploadChunkResponse = await retryHttpClientResponse(
    `uploadChunk (start: ${start}, end: ${end})`,
    async () =>
      httpClient.sendStream(
        'PATCH',
        resourceUrl,
        openStream(),
        additionalHeaders
      )
  )

sendStream return a response, the body of must be consumed to avoid holding the event loop open.

Something like the following ought to do it:

  const uploadChunkResponse = await retryHttpClientResponse(
    `uploadChunk (start: ${start}, end: ${end})`,
    async () => {
      const response = await httpClient.sendStream(
        'PATCH',
        resourceUrl,
        openStream(),
        additionalHeaders
      )

      await response.readBody()

      return response
    }
  )

Start in cache/src/internal/cacheHttpClient at the uploadChunk call to httpClient.sendStream and inspect how its response is handled. Ensure the response body is consumed before returning, then verify that cache uploads no longer leave a socket holding the event loop open.