rawfile-localpv/rawfile_servicer.py
Mehran Kholdi 63c8eb44ba Fix race condition that was causing dangling loop devices
Apparently it is wrong to assume that `DeleteVolume` gets called
only after `UnstageVolume` returns success. This was causing the
disk image file to be deleted while the volume was still mounted.
This would prevent the loop device from getting detached and in
turn disk space from getting reclaimed.
2021-11-19 18:59:49 +03:30

253 lines
8.7 KiB
Python

from pathlib import Path
from subprocess import CalledProcessError
import grpc
from google.protobuf.wrappers_pb2 import BoolValue
import rawfile_util
from consts import (
PROVISIONER_VERSION,
PROVISIONER_NAME,
RESOURCE_EXHAUSTED_EXIT_CODE,
VOLUME_IN_USE_EXIT_CODE,
)
from csi import csi_pb2, csi_pb2_grpc
from declarative import be_symlink, be_absent
from fs_util import device_stats, mountpoint_to_dev
from orchestrator.k8s import volume_to_node, run_on_node
from rawfile_util import attach_loop, detach_loops
from remote import init_rawfile, scrub, get_capacity, expand_rawfile
from util import log_grpc_request, run
NODE_NAME_TOPOLOGY_KEY = "hostname"
class RawFileIdentityServicer(csi_pb2_grpc.IdentityServicer):
@log_grpc_request
def GetPluginInfo(self, request, context):
return csi_pb2.GetPluginInfoResponse(
name=PROVISIONER_NAME, vendor_version=PROVISIONER_VERSION
)
@log_grpc_request
def GetPluginCapabilities(self, request, context):
Cap = csi_pb2.PluginCapability
return csi_pb2.GetPluginCapabilitiesResponse(
capabilities=[
Cap(service=Cap.Service(type=Cap.Service.CONTROLLER_SERVICE)),
Cap(
service=Cap.Service(
type=Cap.Service.VOLUME_ACCESSIBILITY_CONSTRAINTS
)
),
Cap(
volume_expansion=Cap.VolumeExpansion(
type=Cap.VolumeExpansion.ONLINE
)
),
]
)
# @log_grpc_request
def Probe(self, request, context):
return csi_pb2.ProbeResponse(ready=BoolValue(value=True))
class RawFileNodeServicer(csi_pb2_grpc.NodeServicer):
def __init__(self, node_name):
self.node_name = node_name
# @log_grpc_request
def NodeGetCapabilities(self, request, context):
Cap = csi_pb2.NodeServiceCapability
return csi_pb2.NodeGetCapabilitiesResponse(
capabilities=[
Cap(rpc=Cap.RPC(type=Cap.RPC.STAGE_UNSTAGE_VOLUME)),
Cap(rpc=Cap.RPC(type=Cap.RPC.GET_VOLUME_STATS)),
Cap(rpc=Cap.RPC(type=Cap.RPC.EXPAND_VOLUME)),
]
)
@log_grpc_request
def NodePublishVolume(self, request, context):
target_path = request.target_path
staging_path = request.staging_target_path
staging_dev_path = Path(f"{staging_path}/dev")
be_symlink(path=target_path, to=staging_dev_path)
return csi_pb2.NodePublishVolumeResponse()
@log_grpc_request
def NodeUnpublishVolume(self, request, context):
target_path = request.target_path
be_absent(path=target_path)
return csi_pb2.NodeUnpublishVolumeResponse()
@log_grpc_request
def NodeGetInfo(self, request, context):
return csi_pb2.NodeGetInfoResponse(
node_id=self.node_name,
accessible_topology=csi_pb2.Topology(
segments={NODE_NAME_TOPOLOGY_KEY: self.node_name}
),
)
@log_grpc_request
def NodeStageVolume(self, request, context):
img_file = rawfile_util.img_file(request.volume_id)
loop_file = attach_loop(img_file)
staging_path = request.staging_target_path
staging_dev_path = Path(f"{staging_path}/dev")
be_symlink(path=staging_dev_path, to=loop_file)
return csi_pb2.NodeStageVolumeResponse()
@log_grpc_request
def NodeUnstageVolume(self, request, context):
img_file = rawfile_util.img_file(request.volume_id)
staging_path = request.staging_target_path
staging_dev_path = Path(f"{staging_path}/dev")
be_absent(staging_dev_path)
detach_loops(img_file)
return csi_pb2.NodeUnstageVolumeResponse()
# @log_grpc_request
def NodeGetVolumeStats(self, request, context):
volume_path = request.volume_path
dev = mountpoint_to_dev(volume_path)
stats = device_stats(dev=dev)
return csi_pb2.NodeGetVolumeStatsResponse(
usage=[
csi_pb2.VolumeUsage(
total=stats["dev_size"],
unit=csi_pb2.VolumeUsage.Unit.BYTES,
),
]
)
@log_grpc_request
def NodeExpandVolume(self, request, context):
volume_path = request.volume_path
size = request.capacity_range.required_bytes
volume_path = Path(volume_path).resolve()
run(f"losetup -c {volume_path}")
return csi_pb2.NodeExpandVolumeResponse(capacity_bytes=size)
class RawFileControllerServicer(csi_pb2_grpc.ControllerServicer):
@log_grpc_request
def ControllerGetCapabilities(self, request, context):
Cap = csi_pb2.ControllerServiceCapability
return csi_pb2.ControllerGetCapabilitiesResponse(
capabilities=[
Cap(rpc=Cap.RPC(type=Cap.RPC.CREATE_DELETE_VOLUME)),
Cap(rpc=Cap.RPC(type=Cap.RPC.GET_CAPACITY)),
Cap(rpc=Cap.RPC(type=Cap.RPC.EXPAND_VOLUME)),
]
)
@log_grpc_request
def CreateVolume(self, request, context):
# TODO: volume_capabilities
if len(request.volume_capabilities) != 1:
context.abort(
grpc.StatusCode.INVALID_ARGUMENT, "Exactly one cap is supported"
)
volume_capability = request.volume_capabilities[0]
AccessModeEnum = csi_pb2.VolumeCapability.AccessMode.Mode
if volume_capability.access_mode.mode not in [
AccessModeEnum.SINGLE_NODE_WRITER
]:
context.abort(
grpc.StatusCode.INVALID_ARGUMENT,
f"Unsupported access mode: {AccessModeEnum.Name(volume_capability.access_mode.mode)}",
)
# FIXME: re-enable access_type after bd2fs is fixed
# access_type = volume_capability.WhichOneof("access_type")
# if access_type == "block":
# pass
# else:
# context.abort(
# grpc.StatusCode.INVALID_ARGUMENT,
# "PANIC! This should be handled by bd2fs!",
# )
MIN_SIZE = 16 * 1024 * 1024 # 16MiB: can't format xfs with smaller volumes
size = max(MIN_SIZE, request.capacity_range.required_bytes)
try:
node_name = request.accessibility_requirements.preferred[0].segments[
NODE_NAME_TOPOLOGY_KEY
]
except IndexError:
context.abort(
grpc.StatusCode.INVALID_ARGUMENT,
"No preferred topology set. Is external-provisioner running in strict-topology mode?",
)
except KeyError:
context.abort(
grpc.StatusCode.INVALID_ARGUMENT, "Topology key not found... why?"
)
try:
init_rawfile(volume_id=request.name, size=size),
except CalledProcessError as exc:
if exc.returncode == RESOURCE_EXHAUSTED_EXIT_CODE:
context.abort(
grpc.StatusCode.RESOURCE_EXHAUSTED, "Not enough disk space"
)
else:
raise exc
return csi_pb2.CreateVolumeResponse(
volume=csi_pb2.Volume(
volume_id=request.name,
capacity_bytes=size,
accessible_topology=[
csi_pb2.Topology(segments={NODE_NAME_TOPOLOGY_KEY: node_name})
],
)
)
@log_grpc_request
def DeleteVolume(self, request, context):
try:
scrub(volume_id=request.volume_id)
except CalledProcessError as exc:
if exc.returncode == VOLUME_IN_USE_EXIT_CODE:
context.abort(grpc.StatusCode.FAILED_PRECONDITION, "Volume in use")
else:
raise exc
return csi_pb2.DeleteVolumeResponse()
@log_grpc_request
def GetCapacity(self, request, context):
return csi_pb2.GetCapacityResponse(
available_capacity=get_capacity(),
)
@log_grpc_request
def ControllerExpandVolume(self, request, context):
volume_id = request.volume_id
node_name = volume_to_node(volume_id)
size = request.capacity_range.required_bytes
try:
run_on_node(
expand_rawfile.as_cmd(volume_id=volume_id, size=size), node=node_name
)
except CalledProcessError as exc:
if exc.returncode == RESOURCE_EXHAUSTED_EXIT_CODE:
context.abort(
grpc.StatusCode.RESOURCE_EXHAUSTED, "Not enough disk space"
)
else:
raise exc
return csi_pb2.ControllerExpandVolumeResponse(
capacity_bytes=size,
node_expansion_required=True,
)