New RunLLM response generated: >>>*# Question* Hi. I have map tasks that have cardinality N, but concurrency M, where M << N. I am seeing my task stuck in QUEUED state when N is larger than the number of pods available. This is quite unexpected, as M is sufficiently small compared to the pod count. # Answer The issue you're encountering might be related to how Flyte handles concurrency and resource allocation for map tasks. According to Flyte documentation, the

concurrency

parameter limits the number of mapped tasks that can run in parallel. If the input size exceeds the concurrency value, multiple batches will run serially until all inputs are processed. However, if the number of available pods is less than the required concurrency, tasks may get stuck in the

QUEUED

state. Additionally, there are known issues with Flyte's handling of large map tasks and their interaction with Kubernetes, as discussed in this GitHub issue. It mentions that dynamic workflows might be a workaround for large map tasks, but they come with their own overhead and limitations. Would you like more detailed steps on how to troubleshoot or configure your Flyte setup to handle this better? Sources: - Flyte documentation - GitHub issue Slack Link https://flyte-org.slack.com/archives/CP2HDHKE1/p1720651177.648039 👍 0 button 👎 0 button 💡 Hint: Mention @runllm in the thread for followups.