Forum

Hi, I dug into your server (Tencent na-ashburn, 2 vCPU / 2 GB RAM) and the root cause is clear — the node is running out of memory.

• Total RAM: 1.9 GB; currently 1.6 GB used + 613 MB of swap active.
• Memory allocatable to pods: only ~812 MiB.
• You have 4 active projects sharing this single server (v1_cantolyrics_free_b2b, v2_cantolyrics_testing_myself, v2-solfege-to-0243, v3_cantolyrics_cus_paid). The heaviest workload is your v2-solfege backend (~237 MiB).
• The v3-cantolyrics-paid-cus-backend ReplicaSet has hundreds of evicted pods over the past week, all with the same kubelet message: Pod was rejected: The node had condition: [MemoryPressure].
• Your current pod is actually running fine (~30h uptime, 86 MiB). The "stuck in starting" you see is the new replica during a rolling deploy that can't get scheduled because there's no spare RAM to run the old and new pod side-by-side during the rollout.

The fix is to upgrade this server to a larger spec — e.g. 2C/4G or 4C/8G. You can do this from the dashboard's Server page. With 4 GB+ RAM your 4 projects will have headroom and rolling deploys will succeed.

If upgrading isn't an option, alternatives are: split projects across a second server, delete services you no longer need, or accept brief downtime per deploy by setting the strategy to recreate-then-roll.

This post has been inactive for a while. We will be closing it in 2 days if there is no new activity.