@@ -189,6 +189,12 @@ def set_cluster_create_parser(cluster_create_parser: ArgumentParser):
189189 cluster_create_tensorboard_arguments
190190 )
191191
192+ ### MTC arguments specific to "cluster create"
193+ cluster_create_mtc_arguments = cluster_create_parser .add_argument_group (
194+ 'Optional MTC Arguments' ,
195+ 'Arguments for configuring MTC in cluster create.' ,
196+ )
197+ add_shared_cluster_create_mtc_arguments (cluster_create_mtc_arguments )
192198 cluster_create_parser .set_defaults (func = cluster_create )
193199
194200
@@ -244,6 +250,14 @@ def set_cluster_create_pathways_parser(
244250 cluster_create_pathways_tensorboard_arguments
245251 )
246252
253+ ### MTC arguments specific to "cluster create"
254+ cluster_create_mtc_arguments = (
255+ cluster_create_pathways_parser .add_argument_group (
256+ 'Optional MTC Arguments' ,
257+ 'Arguments for configuring MTC in cluster create.' ,
258+ )
259+ )
260+ add_shared_cluster_create_mtc_arguments (cluster_create_mtc_arguments )
247261 cluster_create_pathways_parser .set_defaults (func = cluster_create_pathways )
248262
249263
@@ -313,6 +327,12 @@ def set_cluster_create_ray_parser(cluster_create_ray_parser: ArgumentParser):
313327 cluster_create_ray_tensorboard_arguments
314328 )
315329
330+ ### MTC arguments specific to "cluster create"
331+ cluster_create_mtc_arguments = cluster_create_ray_parser .add_argument_group (
332+ 'Optional MTC Arguments' ,
333+ 'Arguments for configuring MTC in cluster create.' ,
334+ )
335+ add_shared_cluster_create_mtc_arguments (cluster_create_mtc_arguments )
316336 cluster_create_ray_parser .set_defaults (func = cluster_create_ray_cluster )
317337
318338
@@ -706,3 +726,43 @@ def add_shared_cluster_create_capacity_arguments(parser: ArgumentParser):
706726 ' See `--reservation` or `--on-demand` for other capacity types.'
707727 ),
708728 )
729+
730+
731+ def add_shared_cluster_create_mtc_arguments (parser : ArgumentParser ):
732+ """Add shared Multi-tier Checkpointing arguments in cluster create and Pathways cluster create.
733+
734+ Args:
735+ List of cluster create MTC arguments parsers
736+ """
737+ parser .add_argument (
738+ '--enable-mtc' ,
739+ action = 'store_true' ,
740+ help = 'Enable MTC on the cluster.' ,
741+ )
742+ parser .add_argument (
743+ '--mtc-ramdisk-size' ,
744+ type = str ,
745+ default = None ,
746+ help = (
747+ '(Required if --enable-mtc is true) The size of the RAM disk to be'
748+ ' used for multi-tier checkpointing. e.g. "64Mi" '
749+ ),
750+ )
751+ parser .add_argument (
752+ '--mtc-gcs-bucket' ,
753+ type = str ,
754+ default = None ,
755+ help = (
756+ '(Required if --enable-mtc is true) The GCS bucket to be used for'
757+ ' multi-tier checkpointing.'
758+ ),
759+ )
760+ parser .add_argument (
761+ '--mtc-toleration-key' ,
762+ type = str ,
763+ default = None ,
764+ help = (
765+ '(Optional) The tolerance key to be used for multi-tier'
766+ ' checkpointing. By default, it is set to "google.com/tpu".'
767+ ),
768+ )
0 commit comments