parallelvirtualcluster
/
pvc
1
0
Fork 0
Code Issues 11 Pull Requests Projects Releases 111 Wiki Activity

Add timeout for safe-to-destroy 

Continuously take the OSD down and out while doing so.
This commit is contained in:
Joshua Boniface 2023-11-04 00:55:05 -04:00
parent 1535078842
commit 48b2ccbd95
1 changed files with 20 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
node-daemon/pvcnoded/objects/CephInstance.py
View File

@ -592,6 +592,7 @@ class CephOSDInstance(object):
# 2. Wait for the OSD to be safe to remove (but don't wait for rebalancing to complete) # 2. Wait for the OSD to be safe to remove (but don't wait for rebalancing to complete)
logger.out(f"Waiting for OSD {osd_id} to be safe to remove", state="i") logger.out(f"Waiting for OSD {osd_id} to be safe to remove", state="i")
tcount = 0
while True: while True:
retcode, stdout, stderr = common.run_os_command( retcode, stdout, stderr = common.run_os_command(
f"ceph osd safe-to-destroy osd.{osd_id}" f"ceph osd safe-to-destroy osd.{osd_id}"
@ -599,7 +600,16 @@ class CephOSDInstance(object):
if int(retcode) in [0, 11]: if int(retcode) in [0, 11]:
break break
else: else:
common.run_os_command(f"ceph osd down {osd_id}")
common.run_os_command(f"ceph osd out {osd_id}")
time.sleep(1) time.sleep(1)
tcount += 1
if tcount > 60:
logger.out(
f"Timed out (60s) waiting for OSD {osd_id} to be safe to remove; proceeding",
state="w",
)
break
# 3. Stop the OSD process and wait for it to be terminated # 3. Stop the OSD process and wait for it to be terminated
logger.out(f"Stopping OSD {osd_id}", state="i") logger.out(f"Stopping OSD {osd_id}", state="i")
@ -971,6 +981,7 @@ class CephOSDInstance(object):
# 2. Wait for the OSD to be safe to remove (but don't wait for rebalancing to complete) # 2. Wait for the OSD to be safe to remove (but don't wait for rebalancing to complete)
if not force_flag: if not force_flag:
logger.out(f"Waiting for OSD {osd_id} to be safe to remove", state="i") logger.out(f"Waiting for OSD {osd_id} to be safe to remove", state="i")
tcount = 0
while True: while True:
retcode, stdout, stderr = common.run_os_command( retcode, stdout, stderr = common.run_os_command(
f"ceph osd safe-to-destroy osd.{osd_id}" f"ceph osd safe-to-destroy osd.{osd_id}"
@ -978,7 +989,16 @@ class CephOSDInstance(object):
if int(retcode) in [0, 11]: if int(retcode) in [0, 11]:
break break
else: else:
common.run_os_command(f"ceph osd down {osd_id}")
common.run_os_command(f"ceph osd out {osd_id}")
time.sleep(1) time.sleep(1)
tcount += 1
if tcount > 60:
logger.out(
f"Timed out (60s) waiting for OSD {osd_id} to be safe to remove; proceeding",
state="w",
)
break
# 3. Stop the OSD process and wait for it to be terminated # 3. Stop the OSD process and wait for it to be terminated
logger.out(f"Stopping OSD {osd_id}", state="i") logger.out(f"Stopping OSD {osd_id}", state="i")