Re: Wrong %USED and MAX AVAIL stats for pool
The documentation states: https://docs.ceph.com/docs/mimic/rados/operations/monitoring/ The POOLS section of the output provides a list of pools and the notional usage of each pool. The output from this section DOES NOT reflect replicas, clones or snapshots. For example, if you store an object with 1MB of data, the notional usage will be 1MB, but the actual usage may be 2MB or more depending on the number of replicas, clones and snapshots. However in our case we are clearly seeing the USAGE field multiplying the total object sizes to the number of replicas. [root@blackmirror ~]# ceph df RAW STORAGE: CLASS SIZE AVAIL USED RAW USED %RAW USED hdd 80 TiB 34 TiB 46 TiB 46 TiB 58.10 TOTAL 80 TiB 34 TiB 46 TiB 46 TiB 58.10 POOLS: POOL ID STORED OBJECTS USED %USED MAX AVAIL one 2 15 TiB 4.05M 46 TiB 68.32 7.2 TiB bench 5 250 MiB 67 250 MiB 0 22 TiB [root@blackmirror ~]# rbd du -p one NAME PROVISIONED USED ... <TOTAL> 20 TiB 15 TiB This is causing several apps (including ceph dashboard) to display inaccurate percentages, because they calculate the total pool capacity as USED + MAX AVAIL, which in this case yields 53.2TB, which is way off. 7.2TB is about 13% of that, so we receive alarms and this is bugging us for quite some time now.
Hi Yordan, this is mimic documentation and these snippets aren't valid for Nautilus any more. They are still present in Nautilus pages though.. Going to create a corresponding ticket to fix that. Relevant Nautilus changes for 'ceph df [detail]' command can be found in Nautilus release notes: https://docs.ceph.com/docs/nautilus/releases/nautilus/ In short - USED field accounts for all the overhead data including replicas etc. It's STORED field which now represents pure data user put into a pool. Thanks, Igor On 10/2/2019 8:33 AM, Yordan Yordanov (Innologica) wrote:
The documentation states: https://docs.ceph.com/docs/mimic/rados/operations/monitoring/
The *POOLS* section of the output provides a list of pools and the notional usage of each pool. The output from this section *DOES NOT* reflect replicas, clones or snapshots. For example, if you store an object with 1MB of data, the notional usage will be 1MB, but the actual usage may be 2MB or more depending on the number of replicas, clones and snapshots.
However in our case we are clearly seeing the USAGE field multiplying the total object sizes to the number of replicas.
[root@blackmirror ~]# ceph df RAW STORAGE: CLASS SIZE AVAIL USED RAW USED %RAW USED hdd 80 TiB 34 TiB *46 TiB* 46 TiB 58.10 TOTAL 80 TiB 34 TiB 46 TiB 46 TiB 58.10
POOLS: POOL ID STORED OBJECTS USED %USED MAX AVAIL one 2 15 TiB 4.05M *46 TiB* 68.32 7.2 TiB bench 5 250 MiB 67 250 MiB 0 22 TiB
[root@blackmirror ~]# rbd du -p one NAME PROVISIONED USED ... <TOTAL> 20 TiB 15 TiB
This is causing several apps (*including ceph dashboard*) to display inaccurate percentages, because they calculate the total pool capacity as USED + MAX AVAIL, which in this case yields 53.2TB, which is way off. 7.2TB is about 13% of that, so we receive alarms and this is bugging us for quite some time now.
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
Hi Igor, Thank you for responding. In this case this looks like a breaking change. I know of two applications that are now incorrectly displaying the pool usage and capacity, It looks like they both rely on the USED field to be divided by the number of replicas. One of those application is actually the Ceph Dashboard. The other is OpenNebula https://docs.opennebula.org/5.6/deployment/open_cloud_storage_setup/ceph_ds..... See the screenshot from Ceph Dashboard - https://imgur.com/vFFxsti. It is stating that we have used 88% of the available space, because it wrongly assumes that the pool capacity is 47.7TB + 6.7TB = 54.4TB, while it should be more like (47.7TB/3) + 6.7TB = 22.6TB. It's absolutely the same story with our OpenNebula instance - https://imgur.com/MOLbo4g. I'm not sure exactly which update broke this, but it was definitely working correctly before. I looked at OpenNebula's code for ceph datastore monitoring and found that it's parsing the XML output of ceph df --format xml, so it looks like this changed too. From file: /var/lib/one/remotes/tm/ceph/monitor: # ------------ Compute datastore usage ------------- MONITOR_SCRIPT=$(cat <<EOF $CEPH df --format xml EOF ) MONITOR_DATA=$(ssh_monitor_and_log $HOST "$MONITOR_SCRIPT" 2>&1) MONITOR_STATUS=$? if [ "$MONITOR_STATUS" = "0" ]; then XPATH="${DRIVER_PATH}/../../datastore/xpath.rb --stdin" echo -e "$(rbd_df_monitor ${MONITOR_DATA} ${POOL_NAME})" else echo "$MONITOR_DATA" exit $MONITOR_STATUS fi From file: /var/lib/one/remotes/datastore/ceph/ceph_utils.sh #-------------------------------------------------------------------------------- # Parse the output of rbd df in xml format and generates a monitor string for a # Ceph pool. You **MUST** define XPATH util before using this function # @param $1 the xml output of the command # @param $2 the pool name #-------------------------------------------------------------------------------- rbd_df_monitor() { local monitor_data i j xpath_elements pool_name bytes_used free monitor_data=$1 pool_name=$2 while IFS= read -r -d '' element; do xpath_elements[i++]="$element" done < <(echo $monitor_data | $XPATH \ "/stats/pools/pool[name = \"${pool_name}\"]/stats/bytes_used" \ "/stats/pools/pool[name = \"${pool_name}\"]/stats/max_avail") bytes_used="${xpath_elements[j++]:-0}" free="${xpath_elements[j++]:-0}" cat << EOF | tr -d '[:blank:][:space:]' USED_MB=$(($bytes_used / 1024**2))\n TOTAL_MB=$((($bytes_used + $free) / 1024**2))\n FREE_MB=$(($free / 1024**2))\n EOF } I believe Ceph Dashboard is doing the same, because the results are the same. Best Regards, On 7 Oct 2019, at 19:03, Igor Fedotov <ifedotov@suse.de<mailto:ifedotov@suse.de>> wrote: Hi Yordan, this is mimic documentation and these snippets aren't valid for Nautilus any more. They are still present in Nautilus pages though.. Going to create a corresponding ticket to fix that. Relevant Nautilus changes for 'ceph df [detail]' command can be found in Nautilus release notes: https://docs.ceph.com/docs/nautilus/releases/nautilus/ In short - USED field accounts for all the overhead data including replicas etc. It's STORED field which now represents pure data user put into a pool. Thanks, Igor On 10/2/2019 8:33 AM, Yordan Yordanov (Innologica) wrote: The documentation states: https://docs.ceph.com/docs/mimic/rados/operations/monitoring/ The POOLS section of the output provides a list of pools and the notional usage of each pool. The output from this section DOES NOT reflect replicas, clones or snapshots. For example, if you store an object with 1MB of data, the notional usage will be 1MB, but the actual usage may be 2MB or more depending on the number of replicas, clones and snapshots. However in our case we are clearly seeing the USAGE field multiplying the total object sizes to the number of replicas. [root@blackmirror ~]# ceph df RAW STORAGE: CLASS SIZE AVAIL USED RAW USED %RAW USED hdd 80 TiB 34 TiB 46 TiB 46 TiB 58.10 TOTAL 80 TiB 34 TiB 46 TiB 46 TiB 58.10 POOLS: POOL ID STORED OBJECTS USED %USED MAX AVAIL one 2 15 TiB 4.05M 46 TiB 68.32 7.2 TiB bench 5 250 MiB 67 250 MiB 0 22 TiB [root@blackmirror ~]# rbd du -p one NAME PROVISIONED USED ... <TOTAL> 20 TiB 15 TiB This is causing several apps (including ceph dashboard) to display inaccurate percentages, because they calculate the total pool capacity as USED + MAX AVAIL, which in this case yields 53.2TB, which is way off. 7.2TB is about 13% of that, so we receive alarms and this is bugging us for quite some time now. _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io<mailto:ceph-users@ceph.io> To unsubscribe send an email to ceph-users-leave@ceph.io<mailto:ceph-users-leave@ceph.io>
I can confirm that this is a failure in ceph 14.2.4 dashboard - as i am seeing this also when i check the free spare under "pools" Am 8. Oktober 2019 07:54:58 MESZ schrieb "Yordan Yordanov (Innologica)" <Yordan.Yordanov@innologica.com>:
Hi Igor,
Thank you for responding. In this case this looks like a breaking change. I know of two applications that are now incorrectly displaying the pool usage and capacity, It looks like they both rely on the USED field to be divided by the number of replicas. One of those application is actually the Ceph Dashboard. The other is OpenNebula https://docs.opennebula.org/5.6/deployment/open_cloud_storage_setup/ceph_ds..... See the screenshot from Ceph Dashboard - https://imgur.com/vFFxsti. It is stating that we have used 88% of the available space, because it wrongly assumes that the pool capacity is 47.7TB + 6.7TB = 54.4TB, while it should be more like (47.7TB/3) + 6.7TB = 22.6TB. It's absolutely the same story with our OpenNebula instance - https://imgur.com/MOLbo4g. I'm not sure exactly which update broke this, but it was definitely working correctly before. I looked at OpenNebula's code for ceph datastore monitoring and found that it's parsing the XML output of ceph df --format xml, so it looks like this changed too.
From file: /var/lib/one/remotes/tm/ceph/monitor: # ------------ Compute datastore usage -------------
MONITOR_SCRIPT=$(cat <<EOF $CEPH df --format xml EOF )
MONITOR_DATA=$(ssh_monitor_and_log $HOST "$MONITOR_SCRIPT" 2>&1) MONITOR_STATUS=$?
if [ "$MONITOR_STATUS" = "0" ]; then XPATH="${DRIVER_PATH}/../../datastore/xpath.rb --stdin" echo -e "$(rbd_df_monitor ${MONITOR_DATA} ${POOL_NAME})" else echo "$MONITOR_DATA" exit $MONITOR_STATUS fi
From file: /var/lib/one/remotes/datastore/ceph/ceph_utils.sh
#-------------------------------------------------------------------------------- # Parse the output of rbd df in xml format and generates a monitor string for a # Ceph pool. You **MUST** define XPATH util before using this function # @param $1 the xml output of the command # @param $2 the pool name #-------------------------------------------------------------------------------- rbd_df_monitor() {
local monitor_data i j xpath_elements pool_name bytes_used free
monitor_data=$1 pool_name=$2
while IFS= read -r -d '' element; do xpath_elements[i++]="$element" done < <(echo $monitor_data | $XPATH \ "/stats/pools/pool[name = \"${pool_name}\"]/stats/bytes_used" \ "/stats/pools/pool[name = \"${pool_name}\"]/stats/max_avail")
bytes_used="${xpath_elements[j++]:-0}" free="${xpath_elements[j++]:-0}"
cat << EOF | tr -d '[:blank:][:space:]' USED_MB=$(($bytes_used / 1024**2))\n TOTAL_MB=$((($bytes_used + $free) / 1024**2))\n FREE_MB=$(($free / 1024**2))\n EOF }
I believe Ceph Dashboard is doing the same, because the results are the same.
Best Regards,
On 7 Oct 2019, at 19:03, Igor Fedotov <ifedotov@suse.de<mailto:ifedotov@suse.de>> wrote:
Hi Yordan,
this is mimic documentation and these snippets aren't valid for Nautilus any more. They are still present in Nautilus pages though..
Going to create a corresponding ticket to fix that.
Relevant Nautilus changes for 'ceph df [detail]' command can be found in Nautilus release notes: https://docs.ceph.com/docs/nautilus/releases/nautilus/
In short - USED field accounts for all the overhead data including replicas etc. It's STORED field which now represents pure data user put into a pool.
Thanks,
Igor
On 10/2/2019 8:33 AM, Yordan Yordanov (Innologica) wrote: The documentation states: https://docs.ceph.com/docs/mimic/rados/operations/monitoring/
The POOLS section of the output provides a list of pools and the notional usage of each pool. The output from this section DOES NOT reflect replicas, clones or snapshots. For example, if you store an object with 1MB of data, the notional usage will be 1MB, but the actual usage may be 2MB or more depending on the number of replicas, clones and snapshots.
However in our case we are clearly seeing the USAGE field multiplying the total object sizes to the number of replicas.
[root@blackmirror ~]# ceph df RAW STORAGE: CLASS SIZE AVAIL USED RAW USED %RAW USED hdd 80 TiB 34 TiB 46 TiB 46 TiB 58.10 TOTAL 80 TiB 34 TiB 46 TiB 46 TiB 58.10
POOLS: POOL ID STORED OBJECTS USED %USED MAX AVAIL one 2 15 TiB 4.05M 46 TiB 68.32 7.2 TiB bench 5 250 MiB 67 250 MiB 0 22 TiB
[root@blackmirror ~]# rbd du -p one NAME PROVISIONED USED ... <TOTAL> 20 TiB 15 TiB
This is causing several apps (including ceph dashboard) to display inaccurate percentages, because they calculate the total pool capacity as USED + MAX AVAIL, which in this case yields 53.2TB, which is way off. 7.2TB is about 13% of that, so we receive alarms and this is bugging us for quite some time now.
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io<mailto:ceph-users@ceph.io> To unsubscribe send an email to ceph-users-leave@ceph.io<mailto:ceph-users-leave@ceph.io>
I think this might be related to a problem I'm having with "ceph osd pool autoscale-status". SIZE appears to be raw usage (data * 3 in our case) while TARGET SIZE seems to be expecting user-facing size. For example, I have a 87TiB dataset that I'm currently copying into a CephFS. "du -sh" shows that 15TiB of it has been copied so far. SIZE reports 44608G - this is consistent with 15 TiB at size=3. But if I set TARGET SIZE to 87TiB * 3, this pool has a RATIO of over 1 - ie larger than the cluster. I currently have TARGET_SIZE set to 87TiB, and it seems to be working - but it's odd that SIZE and TARGET_SIZE seem to use very different measurements. Does anyone know if this is intended? On Mon, Oct 7, 2019 at 12:03 PM Igor Fedotov <ifedotov@suse.de> wrote:
Hi Yordan,
this is mimic documentation and these snippets aren't valid for Nautilus any more. They are still present in Nautilus pages though..
Going to create a corresponding ticket to fix that.
Relevant Nautilus changes for 'ceph df [detail]' command can be found in Nautilus release notes: https://docs.ceph.com/docs/nautilus/releases/nautilus/
In short - USED field accounts for all the overhead data including replicas etc. It's STORED field which now represents pure data user put into a pool.
Thanks,
Igor
On 10/2/2019 8:33 AM, Yordan Yordanov (Innologica) wrote:
The documentation states: https://docs.ceph.com/docs/mimic/rados/operations/monitoring/
The POOLS section of the output provides a list of pools and the notional usage of each pool. The output from this section DOES NOT reflect replicas, clones or snapshots. For example, if you store an object with 1MB of data, the notional usage will be 1MB, but the actual usage may be 2MB or more depending on the number of replicas, clones and snapshots.
However in our case we are clearly seeing the USAGE field multiplying the total object sizes to the number of replicas.
[root@blackmirror ~]# ceph df RAW STORAGE: CLASS SIZE AVAIL USED RAW USED %RAW USED hdd 80 TiB 34 TiB 46 TiB 46 TiB 58.10 TOTAL 80 TiB 34 TiB 46 TiB 46 TiB 58.10
POOLS: POOL ID STORED OBJECTS USED %USED MAX AVAIL one 2 15 TiB 4.05M 46 TiB 68.32 7.2 TiB bench 5 250 MiB 67 250 MiB 0 22 TiB
[root@blackmirror ~]# rbd du -p one NAME PROVISIONED USED ... <TOTAL> 20 TiB 15 TiB
This is causing several apps (including ceph dashboard) to display inaccurate percentages, because they calculate the total pool capacity as USED + MAX AVAIL, which in this case yields 53.2TB, which is way off. 7.2TB is about 13% of that, so we receive alarms and this is bugging us for quite some time now.
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io
participants (4)
-
ceph＠elchaka.de
-
Igor Fedotov
-
Nathan Fish
-
Yordan Yordanov (Innologica)