Repository navigation
Description
Summary
On the Microsoft Office client (d3590ed6-52b3-4102-aeff-aad2292ab01c), Graph keeps answering some routes with 429, even at very low request rates, while closely related routes work. This breaks chat_members() and affects other calls.
All results below come from one account in one tenant, measured on 2026-11-03 with single unretried requests. The cause is unknown, and it may not apply to other tenants.
Chat routes
Each chat row was tried 3 times on each of a 1:1, a group and a meeting chat, unless noted.
| Route | Result on the Office client |
|---|---|
GET chats/{id}/members |
429, Retry-After: 62, every time, answered in about 5 ms |
GET chats/{id}/messages/{id} |
429, Retry-After: 62 (3 tries, one chat) |
GET me/chats/{id}/messages/{id} |
429, Retry-After: 62 (3 tries, one chat) |
GET chats/{id}/pinnedMessages |
429, Retry-After: 62 (seen on several checks) |
GET chats/{id}?$expand=members |
200, all member ids, 126-256 ms |
GET me/chats/{id}?$expand=members |
200, all member ids, 127-174 ms |
GET me/chats/{id}/messages?$top=1 |
200, 125-155 ms (3 tries, one chat) |
GET me/chats?$top=1 |
200 (once) |
The Teams (1fec8e78) and OfficeHome (4765445b) clients got 403 on chats/{id}/messages/{id} and chats/{id}/pinnedMessages, so switching clients doesn't help on these routes.
Teams' own chat service works, but it's undocumented:
| Route | Result |
|---|---|
GET {chatService}/v1/users/ME/conversations/{id}/messages/{id} |
200, 85-93 ms, attachments included |
GET {chatService}/v1/threads/{id} |
200, 95-474 ms. On a 1:1 it also listed 2 members without 8:orgid: ids, so filter them out |
Directory routes
/me, /users, /organization and /me/memberOf also got 429s on the Office client, but 200 on the Teams and OfficeHome clients for the same user. Since 2.0.0, OutNote moves to the next client when this happens. These 429s came and went during the day, with Retry-After values from 3 to 22 seconds.
What we know about the cause
- It doesn't look like our request volume. The documented directory limit is 3,500-8,000 resource units per 10 seconds per app and tenant, and these tests sent far less. That limit may not be what applies here.
- It looks tied to the client and route, not the user or connection.
- The 429 responses had no
x-ms-throttle-scopeor other throttle headers on the directory routes I checked, so Graph doesn't say which limit applies. - Microsoft says 429s can occur below the published limits, for example under service load or because of a tenant's data volume.
- Not tested: another account or tenant. That would show if this is limited to one tenant.
Impact in OutNote
chat_members()still useschats/{id}/memberswith no fallback, so it fails while this lasts.pinned_messages()uses a throttled route, and the other clients get 403.channel_files()got 429s during the same testing. I haven't looked at which of its calls was refused.chat_message()already avoids this, using the chat service with a Graph fallback, since 2.2.0.- A downstream app couldn't send chat images, because it listed members with this route first.
Workaround today: use note.chat(chat_id)["members"] instead of note.chat_members(chat_id).
Proposed changes
- Make
chat_members()usechats/{id}?$expand=members, keepingchats/{id}/membersas a fallback if needed. Check that the member fields match. - Decide what
pinned_messages()should do, for example a chat service route, if one exists. - Find out which call in
channel_files()is throttled. - Log
x-ms-throttle-*headers on 429s, in case Graph sends them. - Test another tenant to narrow down the cause.
Activity
PianoNic commentedon Nov 4, 2026
chat_members()is fixed in v2.2.1 (#47). The other items are still open.