Decide: what the two halves of the HTTP transport showed when run together

Open
#133 1 comment 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
45/100
Issue type
Documentation
Clarity
Clearly specified
Activity status
Active
Tech stack
django, python

Research direction

Read TRN-14 and ADRs 0011, 0014, 0017, 0027, and 0028, along with issue #132 and the records for HTTP-40 and HTTP-31. Compare the seven listed observations with D2 and step 5's last task (#12, task 8). Done means one short record is proposed and accepted that states each narrowing; the issue says this does not block #128 or #129.

Written by the indexing model from the issue text.

Description

The HTTP connector (#128) and the provider endpoints (#129) were run against each other on one tree before either merged. Calls, replies, each error kind, tokens, the principal, deadlines, the idempotency key and a burst of 20 concurrent calls agreed under WSGI and ASGI. What is left is in the records, which each half follows faithfully:

  1. A request that passed the caller's check can still come back invalid_argument. TRN-14 and ADR 0017 Decision 9 step 2 say a caller that checks its own requests never receives that kind for one. Over HTTP the provider's host can still answer it: a body over its size limit (ADR 0011 row 12, HTTP-40), or plain HTTP where it requires HTTPS (row 5). In-process neither applies, so the same call succeeds there. To weigh: narrow the promise to requests that fail the caller's own check, or give both modes one limit (ADR 0011 open question 5).
  2. A token's lifetime is shared by clock difference and waiting time. A provider accepts a token until exp + 60 on its own clock, so a call with no deadline that waits more than about 120 seconds before authentication is unauthenticated. To weigh: state it in ADR 0027 Decision 4, or choose a longer lifetime within spec 03's cap.
  3. Framing. ADR 0011 Decision 4 lists the headers the host sets and says it sets no other; the endpoint also states Content-Length on every response with a body, which connection reuse depends on.
  4. A repeated Content-Length under ASGI. ADR 0011 says the server joins repeated fields; uvicorn merges equal values and refuses differing ones itself.
  5. HEAD on the well-known paths. Django's WSGI servers drop Content-Length: 0 from a HEAD answer; ADR 0011's sentence follows whatever HTTP-31 says once #129 lands.
  6. The reply body has no bound on the caller's side (ADR 0028 open question 1). To weigh: a bound, and the kind when it is passed.
  7. What a refusal may print. ADR 0014's table prints a refused key and a refused BACKEND value, and a remote provider's url is never printed; #132 found a database password in neolith.E005. To weigh: one rule for every settings refusal.

How: one short record, proposed and accepted with step 5's last task (#12, task 8), stating each narrowing under D2. Nothing here blocks #128 or #129.

Dominant language
Python
Stars
1
Forks
0
Avg merge
5h 30m
Merged PRs (30d)
85

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from stack314/django-neolith

All issues in stack314/django-neolith

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.