WebSocket data feed — sustained 502 Bad Gateway during reconnect, SDK auto-reconnect exhausted, no recovery until manual restart

Automate, API & MCP
,
1

On 14 Aug 2026, my WebSocket connection to the market data feed (fyers_apiv3, FyersDataSocket, reconnect=True) dropped and could not re-establish. Every reconnect attempt received 502 Bad Gateway from Cloudflare (which fronts the WS endpoint), not a response from the origin. The SDK’s built-in reconnect logic exhausted its fixed retry budget and gave up entirely, after which the connection never recovered on its own; we lost live tick data for 7+ minutes until we manually restarted our client.

Timeline (IST, converted from the response headers’ UTC timestamps):

  • 09:55 IST — feed healthy, normal data flowing.
  • ~10:02 IST — WebSocket connection lost (Connection to remote host was lost). SDK’s internal reconnect began.
  • 10:02:23–10:03:05 IST — six consecutive reconnect attempts across two disconnect events, every single one returned 502 Bad Gateway from Cloudflare. Trace IDs (cf-ray) for each attempt:
    • a2ad3c42dd903bb0-BOM
    • a2ad3cca9ce34901-BOM
    • a2ad3ce9ee4e8124-BOM
    • a2ad3d095ff94186-BOM
    • a2ad3d28ce4338b2-BOM
    • a2ad3d482bd846fb-BOM
  • 10:03:05 IST — SDK logged Max reconnect attempts reached. Connection abandoned. No further reconnect was attempted by the SDK after this point.
  • 10:04–10:11+ IST — no ticks received on any subscribed symbol (500+ symbols) for the remainder of this window; recovered only after we manually restarted our ingestion process.

REST API calls (history()) to the same account continued succeeding normally throughout this exact window (09:55, 10:00, 10:05 all returned valid data) so this was isolated to the WebSocket data feed endpoint specifically, not a broader account/auth/network issue on our side.

Was this a known outage on the WS feed infrastructure during this window? Given the response was consistently 502 from Cloudflare rather than a timeout, it looks like the origin server behind Cloudflare was unavailable/unresponsive to Cloudflare itself. I would appreciate confirmation of what happened and whether other API users were similarly affected during this window. It would help me understand if this is a one-off or something to plan resilience around going forward on my side.

Thanks

2

for me it was down for 1min, because i posted, some others did in forum, and it appeared as if their backend restarted their service.

the reconnect logic in the sdk isnt designed for longer breaks timeouts, you should monitor / create some logic to rerun a fresh connection.