d1f8e9b332
FOR UPDATE takes stronger locks, and is only needed if you plan on deleting the row or modifying one of key columns. If you use FOR UPDATE, you block other transactions insert rows that reference a row using a FOR UPDATE lock. From looking at these cases, I don't think that is the case, so using FOR NO KEY UPDATE should allow the assurances we need without unnecessarily blocking other concurrent transactions.
43 lines
1.3 KiB
Ruby
43 lines
1.3 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
require_relative "../model"
|
|
|
|
class VmPool < Sequel::Model
|
|
one_to_one :strand, key: :id
|
|
one_to_many :vms, key: :pool_id
|
|
|
|
plugin ResourceMethods
|
|
|
|
include SemaphoreMethods
|
|
semaphore :destroy
|
|
|
|
def pick_vm
|
|
# Find an available VM in the "running" state and not associated with a GitHub runner,
|
|
# and lock it with FOR NO KEY UPDATE SKIP LOCKED.
|
|
vms_dataset
|
|
.where(Sequel[:vm][:display_state] => "running")
|
|
.exclude(id: DB[:github_runner].exclude(vm_id: nil).select(:vm_id))
|
|
.for_no_key_update
|
|
.skip_locked
|
|
.first
|
|
end
|
|
end
|
|
|
|
# Table: vm_pool
|
|
# Columns:
|
|
# id | uuid | PRIMARY KEY
|
|
# size | integer | NOT NULL
|
|
# vm_size | text | NOT NULL
|
|
# boot_image | text | NOT NULL
|
|
# storage_size_gib | bigint | NOT NULL
|
|
# arch | arch | NOT NULL DEFAULT 'x64'::arch
|
|
# storage_encrypted | boolean | NOT NULL DEFAULT true
|
|
# storage_skip_sync | boolean | NOT NULL DEFAULT false
|
|
# location_id | uuid | NOT NULL
|
|
# Indexes:
|
|
# vm_pool_pkey | PRIMARY KEY btree (id)
|
|
# Foreign key constraints:
|
|
# vm_pool_location_id_fkey | (location_id) REFERENCES location(id)
|
|
# Referenced By:
|
|
# vm | vm_pool_id_fkey | (pool_id) REFERENCES vm_pool(id)
|