秋田预测市场 · Discord 审计

Kalshi API Websocket 连接延迟、溢出及429限流问题

2026-06-13 · Kalshi · 11 条相关讨论

Melon 2026-06-13 00:05:44

Splitting up my subscriptions across multiple websockets seems to have resolved the issue. I have one process using my old code, a it has crashed about 1x per hour on "subscription buffer overflow". The process using the multiple websockets has only crashed once in about 4 hours of running due to no data checks failing (but this occasionally happened to me before so I can't say this is any worse).

Splitting up my subscriptions across multiple websockets seems to have resolved the issue. I have one process using my old code, a it has crashed about 1x per hour on "subscription buffer overflow". The process using the multiple websockets has only crashed once in about 4 hours of running due to no data checks failing (but this occasionally happened to me before so I can't say this is any worse).

Apex 2026-06-13 03:58:07

if you have 1 million messages in your queue and they come in faster than you can process then you can be processing messages that are hours old

if you have 1 million messages in your queue and they come in faster than you can process then you can be processing messages that are hours old

Bumbling Bayesian 2026-06-13 04:03:24

One should definitely queue WS messages and handle them on another thread. But you still need the same amount of compute regardless.

One should definitely queue WS messages and handle them on another thread. But you still need the same amount of compute regardless.

brunoBowser 2026-06-13 05:10:30

Anyone else notice some weird lag issues that started at 5 pm Eastern?

Anyone else notice some weird lag issues that started at 5 pm Eastern?

BayesForDays 2026-06-13 05:38:04

I just had a fill reach me 21 seconds late on the WS 😂 so...yes

I just had a fill reach me 21 seconds late on the WS 😂 so...yes

theed 2026-06-13 05:45:27

getting bad OB data rn

getting bad OB data rn

brunoBowser 2026-06-13 05:46:20

You have been for the last 45 minutes

You have been for the last 45 minutes

theed 2026-06-13 05:46:37

yeah, it stopped for a second, now its bad again

yeah, it stopped for a second, now its bad again

theed 2026-06-13 05:46:54

who pushed to prod on a friday afternoon

who pushed to prod on a friday afternoon

Apex 2026-06-13 06:53:51

ConnectionError: [Errno 9] Bad file descriptor

ConnectionError: [Errno 9] Bad file descriptor

arbGuy 2026-06-13 15:43:08

Hi — I’m seeing a persistent 429 specifically on GET /trade-api/v2/markets. My authenticated GET /account/limits currently reports: * Tier: Basic * Read bucket: 200 capacity / 200 refill rate * Estimated /markets cost: 10 tokens I initially had an overly deep paginated scanner, but I have since reduced it substantially and added defensive backoff: * Normal scan: max 5 pages, 200 markets/page, ~50 estimated tokens * Post-429 probe: exactly 1 page, ~10 estimated tokens * Exponential cooldown: 30m → 60m → 120m → 240m * No retries during cooldown Despite this, even the one-page probe continues to return: 429 {"error":{"code":"too_many_requests","message":"too many requests"}} The rate-limit streak has persisted overnight, while /account/limits and authenticated portfolio endpoints remain available. Does /markets have a separate account-, route-, or IP-level throttle that is not reflected in /account/limits? Is there a longer recovery window after prior high-volume pagination, or anything else I should inspect? I have also applied for Advanced API access, primarily for safer market discovery and batching.

Hi — I’m seeing a persistent 429 specifically on GET /trade-api/v2/markets. My authenticated GET /account/limits currently reports: * Tier: Basic * Read bucket: 200 capacity / 200 refill rate * Estimated /markets cost: 10 tokens I initially had an overly deep paginated scanner, but I have since reduced it substantially and added defensive backoff: * Normal scan: max 5 pages, 200 markets/page, ~50 estimated tokens * Post-429 probe: exactly 1 page, ~10 estimated tokens * Exponential cooldown: 30m → 60m → 120m → 240m * No retries during cooldown Despite this, even the one-page probe continues to return: 429 {"error":{"code":"too_many_requests","message":"too many requests"}} The rate-limit streak has persisted overnight, while /account/limits and authenticated portfolio endpoints remain available. Does /markets have a separate account-, route-, or IP-level throttle that is not reflected in /account/limits? Is there a longer recovery window after prior high-volume pagination, or anything else I should inspect? I have also applied for Advanced API access, primarily for safer market discovery and batching.