From: Peter Amstutz <peter.amstutz@curii.com>
Date: Fri, 7 Jan 2022 22:19:54 +0000 (-0500)
Subject: Merge branch 'main' into 18324-lsf-gpu
X-Git-Tag: 2.4.0~117^2
X-Git-Url: https://git.arvados.org/arvados.git/commitdiff_plain/15f6e58f3b7c158aa735ce78597cf22c6ef3543b?hp=-c

Merge branch 'main' into 18324-lsf-gpu

Arvados-DCO-1.1-Signed-off-by: Peter Amstutz <peter.amstutz@curii.com>
---

15f6e58f3b7c158aa735ce78597cf22c6ef3543b
diff --combined lib/config/config.default.yml
index 002acd03aa,21e39a8a87..17bba5410b
--- a/lib/config/config.default.yml
+++ b/lib/config/config.default.yml
@@@ -899,10 -899,20 +899,20 @@@ Clusters
        # go down.
        MaxComputeVMs: 64
  
-       # Preemptible instance support (e.g. AWS Spot Instances)
-       # When true, child containers will get created with the preemptible
-       # scheduling parameter parameter set.
-       UsePreemptibleInstances: false
+       # Schedule all child containers on preemptible instances (e.g. AWS
+       # Spot Instances) even if not requested by the submitter.
+       #
+       # If false, containers are scheduled on preemptible instances
+       # only when requested by the submitter.
+       #
+       # Note that arvados-cwl-runner does not currently offer a
+       # feature to request preemptible instances, so this value
+       # effectively acts as a cluster-wide decision about whether to
+       # use preemptible instances.
+       #
+       # This flag is ignored if no preemptible instance types are
+       # configured, and has no effect on top-level containers.
+       AlwaysUsePreemptibleInstances: true
  
        # PEM encoded SSH key (RSA, DSA, or ECDSA) used by the
        # cloud dispatcher for executing containers on worker VMs.
@@@ -1089,7 -1099,6 +1099,7 @@@
          # %C number of VCPUs
          # %M memory in MB
          # %T tmp in MB
 +        # %G number of GPU devices (runtime_constraints.cuda.device_count)
          #
          # Use %% to express a literal %. The %%J in the default will be changed
          # to %J, which is interpreted by bsub itself.
@@@ -1100,11 -1109,6 +1110,11 @@@
          # from /tmp, or adjust the "-o" and "-e" arguments accordingly.
          BsubArgumentsList: ["-o", "/tmp/crunch-run.%%J.out", "-e", "/tmp/crunch-run.%%J.err", "-J", "%U", "-n", "%C", "-D", "%MMB", "-R", "rusage[mem=%MMB:tmp=%TMB] span[hosts=1]", "-R", "select[mem>=%MMB]", "-R", "select[tmp>=%TMB]", "-R", "select[ncpus>=%C]"]
  
 +        # Arguments that will be appended to the bsub command line
 +        # when submitting Arvados containers as LSF jobs with
 +        # runtime_constraints.cuda.device_count > 0
 +        BsubCUDAArguments: ["-gpu", "num=%G"]
 +
          # Use sudo to switch to this user account when submitting LSF
          # jobs.
          #
diff --combined sdk/go/arvados/config.go
index e367c9202c,a1ab713e4a..b8c8269f12
--- a/sdk/go/arvados/config.go
+++ b/sdk/go/arvados/config.go
@@@ -431,23 -431,23 +431,23 @@@ type InstanceType struct 
  }
  
  type ContainersConfig struct {
- 	CloudVMs                    CloudVMsConfig
- 	CrunchRunCommand            string
- 	CrunchRunArgumentsList      []string
- 	DefaultKeepCacheRAM         ByteSize
- 	DispatchPrivateKey          string
- 	LogReuseDecisions           bool
- 	MaxComputeVMs               int
- 	MaxDispatchAttempts         int
- 	MaxRetryAttempts            int
- 	MinRetryPeriod              Duration
- 	ReserveExtraRAM             ByteSize
- 	StaleLockTimeout            Duration
- 	SupportedDockerImageFormats StringSet
- 	UsePreemptibleInstances     bool
- 	RuntimeEngine               string
- 	LocalKeepBlobBuffersPerVCPU int
- 	LocalKeepLogsToContainerLog string
+ 	CloudVMs                      CloudVMsConfig
+ 	CrunchRunCommand              string
+ 	CrunchRunArgumentsList        []string
+ 	DefaultKeepCacheRAM           ByteSize
+ 	DispatchPrivateKey            string
+ 	LogReuseDecisions             bool
+ 	MaxComputeVMs                 int
+ 	MaxDispatchAttempts           int
+ 	MaxRetryAttempts              int
+ 	MinRetryPeriod                Duration
+ 	ReserveExtraRAM               ByteSize
+ 	StaleLockTimeout              Duration
+ 	SupportedDockerImageFormats   StringSet
+ 	AlwaysUsePreemptibleInstances bool
+ 	RuntimeEngine                 string
+ 	LocalKeepBlobBuffersPerVCPU   int
+ 	LocalKeepLogsToContainerLog   string
  
  	JobsAPI struct {
  		Enable         string
@@@ -486,7 -486,6 +486,7 @@@
  	LSF struct {
  		BsubSudoUser      string
  		BsubArgumentsList []string
 +		BsubCUDAArguments []string
  	}
  }