-
Notifications
You must be signed in to change notification settings - Fork 129
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Do not wait for slot initialization during exec node start #588
Labels
mapreduce
MapReduce related
Comments
abogutskiy
added a commit
to abogutskiy/ytsaurus
that referenced
this issue
Jun 5, 2024
Resolves ytsaurus#588
abogutskiy
added a commit
to abogutskiy/ytsaurus
that referenced
this issue
Jun 6, 2024
Resolves ytsaurus#588
abogutskiy
added a commit
to abogutskiy/ytsaurus
that referenced
this issue
Jun 10, 2024
Resolves ytsaurus#588
abogutskiy
added a commit
to abogutskiy/ytsaurus
that referenced
this issue
Jun 10, 2024
Resolves ytsaurus#588
abogutskiy
added a commit
to abogutskiy/ytsaurus
that referenced
this issue
Jun 19, 2024
Resolves ytsaurus#588
abogutskiy
added a commit
to abogutskiy/ytsaurus
that referenced
this issue
Jun 21, 2024
Resolves ytsaurus#588
abogutskiy
added a commit
to abogutskiy/ytsaurus
that referenced
this issue
Jun 24, 2024
Resolves ytsaurus#588
abogutskiy
added a commit
to abogutskiy/ytsaurus
that referenced
this issue
Jul 11, 2024
Resolves ytsaurus#588
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Slot is an exec node entity that is used for a single job execution. For example, the number of slots on the exec node limits the maximum number of concurrent jobs on it.
Slot manager (https://github.com/ytsaurus/ytsaurus/blob/main/yt/yt/server/node/exec_node/slot_manager.cpp) is an entity responsible for slot management: it initializes slots when exec node starts and tracks the free slots when jobs start and finish.
When node starts, slot manager initializes all the slots. In CRI job environment in creates slot_count pods which may be slot, since typical slot_count is 4*cpu_count. During the initialization, node reports to scheduler that there are 0 free slots, so no jobs are executed.
This ticket proposes to make initialization one-by-one and async. Slot manager should start slots one-by-one and when a particular slot is ready it becomes accounted in the slot_count reported by scheduler, so it starts to run jobs on the exec node.
https://github.com/ytsaurus/ytsaurus/blob/main/yt/yt/server/node/exec_node/slot_manager.cpp#L247-L250 -- this is the long call that should be split into the number of smaller calls. I propose to split it into Init and InitSlot methods with first initializing job environment without slots and the second one initializes particular slot.
The text was updated successfully, but these errors were encountered: