a month ago
-
Service: airtel-bulk-operation.up.railway.app (Python 3.13.14, US West, 1 replica) - app becomes completely unreachable (page hangs, never loads) at random intervals, with no errors in logs.
-
Pattern: works fine after a manual Restart/Redeploy, then degrades again after a period of use - not a one-time crash, it’s recurring.
-
Observed CPU pegs at 1.0–1.5 vCPU continuously (not idling near 0%) and memory climbs to ~3 GB during the degraded window, then drops back down after restart.
-
Error Logs and Request Error Rate show 0% errors throughout - so the app isn’t crashing/throwing exceptions, it appears to be getting stuck/unresponsive instead.
-
Requesting help identifying whether this is a resource-limit issue (memory/CPU ceiling on current plan) or a platform-side issue (health check, container scheduling), since app-side logs show no explicit failure.
5 Replies
a month ago
This thread has been opened as a bounty so the community can help solve it.
Status changed to Open Railway • about 1 month ago
a month ago
The sustained CPU usage and rising memory suggest repeated computation or accumulating resources, but those metrics alone don’t confirm an app issue or a Railway limit.
Railway’s healthchecks only run during deployment, and restart policies only act when the process exits, so a running but stuck app can remain unresponsive.
Could you share your Streamlit version, start command, configured replica CPU/memory limits, and the code handling bulk operations, caching, or background workers?
During the next hang, before restarting, capture Python thread stacks or a short CPU profile and check whether the app responds locally inside the container. Local failure would point toward the process/container; local success while the public URL hangs would help narrow the investigation toward networking.
I’d first check for bulk jobs repeating on Streamlit reruns, retry loops without delays/timeouts, unbounded caches, and threads or subprocesses accumulating across sessions. The 0% error rate doesn’t rule out these kinds of hangs.
a month ago
sorry about that how about u
a month ago
This is most likely an app-level problem rather than a platform one. CPU pinned and memory climbing with 0% errors suggests the process is busy or leaking, not crashing.
• Railway's billing docs describe a leaking app as one that will "gradually consume more RAM over time."
• Healthchecks only run when a deployment starts, so they won't catch a hang.
• The restart policy only acts when the service stops or exits.
• 3 GB looks well below Hobby limits per the plans page. Check the replica limits in service settings to be sure.To narrow it down: what does the bulk operation do, and how large are the inputs? Do you use st.cache_data or st.cache_resource, or store dataframes in st.session_state? (Streamlit-specific, not from Railway docs.)
Disclosure: drafted with AI assistance and reviewed by me.
sundayabasi684-byte
This is most likely an app-level problem rather than a platform one. CPU pinned and memory climbing with 0% errors suggests the process is busy or leaking, not crashing. • Railway's billing docs describe a leaking app as one that will "gradually consume more RAM over time." • Healthchecks only run when a deployment starts, so they won't catch a hang. • The restart policy only acts when the service stops or exits. • 3 GB looks well below Hobby limits per the plans page. Check the replica limits in service settings to be sure. To narrow it down: what does the bulk operation do, and how large are the inputs? Do you use st.cache_data or st.cache_resource, or store dataframes in st.session_state? (Streamlit-specific, not from Railway docs.) Disclosure: drafted with AI assistance and reviewed by me.
a month ago
The system is down
a month ago
What about now