Details
-
Improvement
-
Resolution: Duplicate
-
Minor
-
Lustre 2.7.0, Lustre 2.5.3, Lustre 2.8.0
-
None
-
9223372036854775807
Description
The lfs df output reports inactive OSTs:
lfs df UUID 1K-blocks Used Available Use% Mounted on myth-MDT0000_UUID 9174328 629228 8020812 7% /myth[MDT:0] myth-OST0000_UUID 3880285544 3431022920 255223648 93% /myth[OST:0] myth-OST0001_UUID 3886052008 2209970436 1481768956 60% /myth[OST:1] myth-OST0002_UUID 3880285544 3554012112 209851112 94% /myth[OST:2] OST0003 : inactive device myth-OST0004_UUID 3880285544 3688294444 191991100 95% /myth[OST:4] filesystem summary: 15526908640 12883299912 2138834816 86% /myth
However, if the OST is permanently disabled vs. just temporarily deactivated, it would be better not to list them at all as this has been reported several times as annoying to users.
One option is to return a different error code for OSTs that have been permanently deactivated vs. ones temporarily deactivated via "lctl --device X deactivate". I see in lov_iocontrol()::IOC_OBD_STATFS that devices that have never been configured (lov->lov_tgts[index] == NULL) return -EAGAIN and that skips all output, vs. inactive devices (lov->lov_tgts[index]->ltd_active == 0) return -ENODATA.
Attachments
Issue Links
Activity
| Fix Version/s | New: Lustre 2.10.0 [ 12204 ] | |
| Resolution | New: Duplicate [ 3 ] | |
| Status | Original: Open [ 1 ] | New: Resolved [ 5 ] |
| Description |
Original:
The {{lfs df}} output reports inactive OSTs: {noformat} lfs df UUID 1K-blocks Used Available Use% Mounted on myth-MDT0000_UUID 9174328 629228 8020812 7% /myth[MDT:0] myth-OST0000_UUID 3880285544 3431022920 255223648 93% /myth[OST:0] myth-OST0001_UUID 3886052008 2209970436 1481768956 60% /myth[OST:1] myth-OST0002_UUID 3880285544 3554012112 209851112 94% /myth[OST:2] OST0003 : inactive device myth-OST0004_UUID 3880285544 3688294444 191991100 95% /myth[OST:4] filesystem summary: 15526908640 12883299912 2138834816 86% /myth {noformat} However, if the OST is permanently disabled vs. just temporarily deactivated, it would be better not to list them at all as this has been reported several times as annoying to users. One option is to return a different error code for OSTs that have been permanently deactivated vs. ones temporarily deactivated via "lctl --device X deactivate". I see in {{lov_iocontrol()::IOC_OBD_STATFS}} that devices that have never been configured (lov->lov_tgts[index] == NULL) return {{-EAGAIN}} and that skips all output, vs. inactive devices (lov->lov_tgts[index]->ltd_active == 0) return {{-ENODATA}}. |
New:
The {{lfs df}} output reports inactive OSTs: {noformat} lfs df UUID 1K-blocks Used Available Use% Mounted on myth-MDT0000_UUID 9174328 629228 8020812 7% /myth[MDT:0] myth-OST0000_UUID 3880285544 3431022920 255223648 93% /myth[OST:0] myth-OST0001_UUID 3886052008 2209970436 1481768956 60% /myth[OST:1] myth-OST0002_UUID 3880285544 3554012112 209851112 94% /myth[OST:2] OST0003 : inactive device myth-OST0004_UUID 3880285544 3688294444 191991100 95% /myth[OST:4] filesystem summary: 15526908640 12883299912 2138834816 86% /myth {noformat} However, if the OST is permanently disabled vs. just temporarily deactivated, it would be better not to list them at all as this has been reported several times as annoying to users. One option is to return a different error code for OSTs that have been permanently deactivated vs. ones temporarily deactivated via "lctl \-\-device X deactivate". I see in {{lov_iocontrol()::IOC_OBD_STATFS}} that devices that have never been configured (lov\->lov_tgts[index] == NULL) return {{\-EAGAIN}} and that skips all output, vs. inactive devices (lov\->lov_tgts[index]\->ltd_active == 0) return {{-ENODATA}}. |
| Link | New: This issue is related to BOS-8 [ BOS-8 ] |
This was fixed via
LU-8920for the 2.10 release.LU-8920 utils: don't print deactivated OSTs in "lfs df" Don't print deactivated OSTs or MDTs in "lfs df". If they have been deactivated then the sysadmin knows they are inactive so there isn't much benefit to printing them all the time. Only print deactivated OSTs/MDTs if the new "-v" option is given. Add long options for "-i" and "-h" (--inodes and --human-readable, respectively) and sort the lfs_df() options to be alphabetical by short option letter. Move the "lfs df" manual to a separate "lfs-df.1" man page to allow better description of the options and examples. Add a "flags" argument to showdf() and mntdf() instead of adding an every-growing list of boolean arguments to these functions. Reviewed-on: https://review.whamcloud.com/24228