From 698fd9e9ba16e107cf3f6b9c9bc28a3882b1eea7 Mon Sep 17 00:00:00 2001
From: Lucas Nussbaum <lucas.nussbaum@loria.fr>
Date: Fri, 22 Sep 2023 17:58:37 +0200
Subject: [PATCH] [gen:oar-prop] Deal with adding nodes to clusters (#14865)
This commit modifies how the generator deals with cases where new
resources are added to an existing cluster, which was not supported up
to now.
The same code is now used for adding nodes to an existing cluster, and
for adding a completely new cluster.
---
lib/refrepo/gen/oar-properties.rb | 53 ++++++++++++-------------------
1 file changed, 20 insertions(+), 33 deletions(-)
diff --git a/lib/refrepo/gen/oar-properties.rb b/lib/refrepo/gen/oar-properties.rb
index 86bf28b7d3..27a1205763 100644
--- a/lib/refrepo/gen/oar-properties.rb
+++ b/lib/refrepo/gen/oar-properties.rb
@@ -1174,9 +1174,6 @@ def extract_clusters_description(clusters, site_name, options, data_hierarchy, s
# (2-b) Detect existing resource_ids and {CPU, CORE, CPUSET, GPU}'s IDs
############################################
- # Detect if the cluster is new, or if it is already known by OAR
- is_a_new_cluster = cluster_resources.select{|x| x["cluster"] == cluster_name}.length == 0
-
# <phys_rsc_map> is a hash that centralises variables that will be used for managing IDs of CPUs, COREs, GPUs of
# the cluster that is going to be updated. <phys_rsc_map> is useful to detect situations where the current number
# of resources associated to a cluster does not correspond to the needs of the cluster.
@@ -1199,47 +1196,37 @@ def extract_clusters_description(clusters, site_name, options, data_hierarchy, s
}
# For each physical ressource, we prepare a list of IDs:
- # a) if the cluster is new: the IDs is a list of number in [max_resource_id, max_resource_id + cluster_resource_count]
- # a) if the cluster is not new: the IDs is the list of existing resources
+ # a) we start with the IDs of the list of existing resources
+ # b) we add more if needed
phys_rsc_map.each do |physical_resource, variables|
- # if it's a new cluster, or the cluster doesn't have resource ids for this kind of resources
- if is_a_new_cluster or cluster_resources.map{|r| r[physical_resource]}.select{|x| not x.nil?}.empty?
- variables[:current_ids] = [*next_rsc_ids[physical_resource]+1..next_rsc_ids[physical_resource]+variables[:per_cluster_count]]
+ current_ids = cluster_resources.map{|r| r[physical_resource]}.select{|x| not x.nil?}.uniq
+ if current_ids.length > variables[:per_cluster_count]
+ raise "#{physical_resource}: too many IDs in OAR compared to what we need to generate. This case needs to be fixed manually."
+ elsif current_ids.length < variables[:per_cluster_count]
+ # there are less resources currently affected to that cluster than what is needed. affect more and bump next_rsc_ids
+ needed = variables[:per_cluster_count] - current_ids.length
+ current_ids += [*next_rsc_ids[physical_resource]+1..next_rsc_ids[physical_resource]+needed]
next_rsc_ids[physical_resource] = variables[:per_server_count] > 0 ? variables[:current_ids].max : next_rsc_ids[physical_resource]
else
- variables[:current_ids] = cluster_resources.map{|r| r[physical_resource]}.select{|x| not x.nil?}.uniq
+ # that's OK, nothing to do
end
+ variables[:current_ids] = current_ids
end
- if is_a_new_cluster
- oar_resource_ids = phys_rsc_map["core"][:current_ids].map{|_r| -1}
- else
- oar_resource_ids = cluster_resources.map{|r| r["id"]}.uniq
- end
-
- phys_rsc_map.each do |physical_resource, variables|
- # Try to detect case where an existing cluster is misconfigured: too many or too few OAR resources
- phys_rsc_ids = variables[:current_ids]
- expected_phys_rsc_count = variables[:per_cluster_count]
-
- if phys_rsc_ids.length != expected_phys_rsc_count
- if ["cpu", "core", "gpu"].include? physical_resource
- puts("#{physical_resource.upcase} has an unexpected number of resources (current:#{phys_rsc_ids.length} vs expected:#{expected_phys_rsc_count}).")
- if ["cpu", "core"].include? physical_resource # this problem is not fatal for GPUs
- raise "unexpected number (current:#{phys_rsc_ids.length} vs expected:#{expected_phys_rsc_count}) of resources for cluster #{cluster_name}"
- end
- end
- end
-
- variables[:current_ids] = phys_rsc_ids
+ oar_resource_ids = cluster_resources.map{|r| r["id"]}.uniq
+ if oar_resource_ids.length < phys_rsc_map["core"][:current_ids].length
+ # the cluster has less resource ids than needed (because resources were added to it)
+ needed = phys_rsc_map["core"][:current_ids].length - oar_resource_ids.length
+ # -1 here asks to generate the resources, not add them
+ oar_resource_ids += [*1..needed].map { -1 }
end
# Some cluster (econome) have attributed resources according to the "alpha-numerical" order of nodes
# ([1, 11, 12, ..., 3] instead of [1, 2, 3, 4, ...]). Here we preserve to order of existing nodes of the cluster
- if is_a_new_cluster
+ nodes_names = cluster_resources.map{|r| r["host"]}.map{|fqdn| {:fqdn => fqdn, :name => fqdn.split(".")[0]}}.uniq
+ if nodes_names.length != node_count
+ # some missing nodes. we regenerate the list from scratch
nodes_names = (1..node_count).map {|i| {:name => "#{cluster_name}-#{i}", :fqdn => "#{cluster_name}-#{i}.#{site_name}.grid5000.fr"}}
- else
- nodes_names = cluster_resources.map{|r| r["host"]}.map{|fqdn| {:fqdn => fqdn, :name => fqdn.split(".")[0]}}.uniq
end
############################################
--
GitLab