New Edge Proxy 5xx errors
All our projects using the new edge proxy are having many intermittent 5xx errors since the 1st of august.
I guess there have been some updates to the proxy resulting in these errors.
Our projects often consists of a Strapi CMS and/or a MedusaJS backend with a NextJS frontend. If the backends use the Edge proxy these errors will directly occur during build time of the NextJS project, but also during runtimes when a dynamic fetch is done by the NextJS server to one of the backends.
Switching back to the legacy proxy works perfectly fine.
6 Replies
Project ID:
fc130f30-4c71-4ab5-99b3-dbff85ca4deafc130f30-4c71-4ab5-99b3-dbff85ca4dea
You can see for example these production error logs where since 1 aug many errors occur:
- https://railway.app/project/fc130f30-4c71-4ab5-99b3-dbff85ca4dea/logs?filter=%40level%3Aerror&start=1721117351871
- https://railway.app/project/3160979a-66a1-4ab0-b77f-1ef3fe72acf4/logs?filter=%40level%3Aerror&start=1721117642017
I have added a reproduction staging environment for this project: https://railway.app/project/fc130f30-4c71-4ab5-99b3-dbff85ca4dea
Building the NextJs Frontend when having the new Edge Proxy enabled on the Strapi CMS will result in build failures where fetches to the CMS fails. Switching on Legacy proxy results in succsfull builds.
Adding some verbose logging gives these fetch error:
SocketError: other side closed / UND_ERR_SOCKET
I have downgraded the Node version on the NextJs project to v18 (was v20) and now the fetches seem to work
Seems weird that this issue showed up on the new Edge Proxy, but maybe the new proxy has some requirements on which HTTP version it supports.
Yeah, i feel it too, somehow after the last railway update, my nextjs always ran into an unknown error, it's ok before
Hey, those two log links don't have a defined range, but defining a range such as Aug 01 - Aug - 02 does show 52(2/4/5) errors.
https://railway.app/project/fc130f30-4c71-4ab5-99b3-dbff85ca4dea/logs?filter=%40level%3Aerror+Request+failed+with+status%3A&start=1722528000000&end=1722614400000
I have added a reproduction staging environment for this project: https://railway.app/project/fc130f30-4c71-4ab5-99b3-dbff85ca4dea Building the NextJs Frontend when having the new Edge Proxy enabled on the Strapi CMS will result in build failures where fetches to the CMS fails. Switching on Legacy proxy results in succsfull builds.the cms in staging has the legacy proxy off, but has successfully deployed, are these errors sporadic? unfortunately an unknown error isn't too helpful in situations like these, i would appreciate if you could open your own help thread and provide a similar or more amount of information as Pepijn has.
Hi Brody, thanks for your replies. This is maybe a better view: https://railway.app/project/fc130f30-4c71-4ab5-99b3-dbff85ca4dea/logs?filter=%40level%3Aerror&start=1722463200000&end=1722852000000 You can see the errors in the production environment
In staging I downgraded the NextJs Node version to v18 and from that point on the fetches worked again.
I just created a new node v20 instance, not the fetches seem to work again
So not really sure if it was maybe related to any of the issues from this incident https://discord.com/channels/713503345364697088/846875565357006878/1270411636527726674
the incidents impacting the edge proxy, yes.
at this time there have been no issues in the reporting for the edge proxy probes for the last 12 hours, so please let us know if you have any issues going forward!