NVIDIA AIStore on Zoned Storage
NVIDIA AIStore (AIS) is a scale-out object storage system. A cluster consists of proxies (gateways), which serve the API and own the cluster map, and targets, which store objects. Each target stores objects as regular files, with per-object metadata in extended attributes, on a set of mountpaths. A mountpath is a directory on a local file system. AIS requires each mountpath to be a separate file system, and recommends XFS.
Format each host-managed SMR disk or NVMe ZNS namespace with zoned XFS and configure each file system as one target mountpath. Object contents go to the sequential zones. XFS metadata, including the extended attributes that hold AIS object metadata, goes to the disk's conventional zones (SMR) or a separate conventional device (ZNS). The proxy and the target's own configuration directory stay on conventional storage. AIS uses standard file I/O and needs no zoned-device support.
- Linux kernel: 6.15+ with CONFIG_XFS_FS and CONFIG_XFS_RT enabled
- xfsprogs: 6.15+
- AIStore 5.2 (release tag v1.5.2): aisnode built from source with Go 1.27.3, ais CLI 1.34
- Tested on Linux 6.17.7 with xfsprogs 6.19.0, with one proxy and one target with four mountpaths on 16 GiB zloop devices
How It Fits Together
| Component | Stores | Placement |
|---|---|---|
| Proxy (aisnode -role=proxy) | Cluster map, bucket metadata, cluster configuration in its confdir | Conventional storage (SSD) |
| Target (aisnode -role=target) | Configuration and copies of cluster metadata in its confdir | Conventional storage (SSD) |
| Target mountpaths | Objects under <mountpath>/@ais/<bucket>/%ob/, work files under %wk/, small metadata files (.ais.vmd, .ais.bmd, .ais.smap) at the mountpath root | One zoned XFS file system per drive |
| Object metadata (user.ais.lom xattr) | Checksum, version, size, copy locations | XFS metadata: conventional zones or ZNS metadata device |
conventional SSD zoned drives (XFS)
+----------------------------+ +-------------+ +-------------+
proxy | /var/lib/ais/proxy | | /data/disk1 | ... | /data/disk4 |
+----------------------------+ | @ais/... | | @ais/... |
+----------------------------+ +-------------+ +-------------+
target | /var/lib/ais/target | target mountpaths (fspaths)
+----------------------------+
A target distributes objects across its mountpaths by object name. With mirroring enabled on a bucket, a target keeps additional copies of each object on other mountpaths. Erasure coding distributes slices across targets, not mountpaths.
Quick Start
These steps set up one proxy and one target with four zoned drives on one
host. Commands prefixed with # run as root. Commands prefixed with $ run
as the ais account created in Step 1, in a login shell (sudo -iu ais),
which builds AIS in its home directory and owns the mountpaths.
Step 1: Prepare Zoned XFS File Systems
Check the XFS requirements before formatting the drives.
For SMR disks (or any zoned device with conventional zones at the start of its address space), format each whole disk with a label:
# mkfs.xfs -L AISDISK1 /dev/sda
meta-data=/dev/sda isize=512 agcount=4, agsize=65536 blks
= sectsz=4096 attr=2, projid32bit=1
= crc=1 finobt=1, sparse=1, rmapbt=1
= reflink=0 bigtime=1 inobtcount=1 nrext64=1
= exchange=1 metadir=1
data = bsize=4096 blocks=262144, imaxpct=25
= sunit=0 swidth=0 blks
naming =version 2 bsize=4096 ascii-ci=0, ftype=1, parent=1
log =internal log bsize=4096 blocks=25087, version=2
= sectsz=4096 sunit=1 blks, lazy-count=1
realtime =internal extsz=4096 blocks=3932160, rtextents=3932160
= rgcount=60 rgsize=65536 extents
= zoned=1 start=262144 reserved=0
Resetting zones...Done.
# mkfs.xfs -L AISDISK2 /dev/sdb
# mkfs.xfs -L AISDISK3 /dev/sdc
# mkfs.xfs -L AISDISK4 /dev/sdd
zoned=1 in the realtime section confirms the zoned realtime device. Create
the account that builds and runs AIS. The build in Step 2 needs a home
directory and Go in its PATH:
# useradd --create-home --user-group --shell /bin/bash ais
Mount the file systems with noatime, which the AIS performance guide
recommends, and give the ais account ownership:
# mkdir -p /data/disk{1..4}
# for i in 1 2 3 4; do mount -o noatime LABEL=AISDISK$i /data/disk$i; done
# chown ais:ais /data/disk{1..4}
# findmnt -t xfs -o SOURCE,TARGET,OPTIONS
SOURCE TARGET OPTIONS
/dev/sda /data/disk1 rw,noatime,attr2,inode64,logbufs=8,logbsize=32k,noquota,max_open_zones=15
/dev/sdb /data/disk2 rw,noatime,attr2,inode64,logbufs=8,logbsize=32k,noquota,max_open_zones=15
/dev/sdc /data/disk3 rw,noatime,attr2,inode64,logbufs=8,logbsize=32k,noquota,max_open_zones=15
/dev/sdd /data/disk4 rw,noatime,attr2,inode64,logbufs=8,logbsize=32k,noquota,max_open_zones=15
The kernel log reports the zone count and open-zone limit for each file system:
XFS (sda): EXPERIMENTAL zoned RT device feature enabled. Use at your own risk!
XFS (sda): Mounting V5 Filesystem a0b461b9-937d-4da9-84fa-bee945f1c97e
XFS (sda): Ending clean mount
XFS (sda): 60 zones of 65536 blocks size (15 max open)
An NVMe ZNS namespace has no conventional zones. Place the XFS metadata on a conventional block device, such as a conventional namespace or an SSD partition, and specify the ZNS namespace as the realtime device:
# mkfs.xfs -L AISDISK1 -r rtdev=/dev/nvme0n2 /dev/nvme1n1
# mount -o noatime,rtdev=/dev/nvme0n2 LABEL=AISDISK1 /data/disk1
The ZNS commands were not tested for this guide. See XFS for details on both cases.
To try this setup without zoned drives, create emulated devices with zloop as described in Zoned Block Device Emulation and use them in place of /dev/sda to /dev/sdd. This guide was tested with four 16 GiB zloop devices with 256 MiB zones and four conventional zones each.
To mount the file systems at boot, add one /etc/fstab entry per drive (not tested on the test host):
# SMR HDDs
LABEL=AISDISK1 /data/disk1 xfs defaults,noatime 0 2
LABEL=AISDISK2 /data/disk2 xfs defaults,noatime 0 2
LABEL=AISDISK3 /data/disk3 xfs defaults,noatime 0 2
LABEL=AISDISK4 /data/disk4 xfs defaults,noatime 0 2
# NVMe ZNS SSD with a conventional metadata device
LABEL=AISDISK1 /data/disk1 xfs defaults,noatime,rtdev=/dev/nvme0n2 0 2
NVMe device names can change across reboots. Prefer a /dev/disk/by-id/nvme-... name for the rtdev= option.
Step 2: Build AIStore
The v1.5.2 release assets contain the ais CLI (ais-linux-amd64.tar.gz) but not the aisnode server binary. Build both from the release tag. The go.mod file requires Go 1.27; if your installed Go is older, select a toolchain explicitly:
$ git clone --depth 1 -b v1.5.2 https://github.com/NVIDIA/aistore.git
$ cd aistore
$ GOTOOLCHAIN=go1.27.3 make node cli
Building aisnode 38dfeb3 [build tags: mono]
...
*** To enable autocompletions in your current shell, run:
*** 'source cmd/cli/autocomplete/bash' or 'source cmd/cli/autocomplete/zsh'
$ ~/go/bin/aisnode -h 2>&1 | grep Version
Version 5.2.38dfeb3 (build: 2026-11-03T16:05:50+0100)
$ ~/go/bin/ais version
version 1.34.38dfeb3 (build 2026-11-03T16:06:47+0100)
make writes the binaries to $GOBIN, or $GOPATH/bin (by default
~/go/bin of the ais account). Install them in /usr/local/bin:
# install -m 755 ~ais/go/bin/aisnode ~ais/go/bin/ais /usr/local/bin/
Step 3: Configure AIStore
Each node reads two files at startup: the initial cluster configuration
(-config) and its local configuration (-local_config). Create the
directories for configuration and logs on conventional storage:
# mkdir -p /etc/ais /var/lib/ais/{proxy,target} /var/log/ais/{proxy,target}
# chown -R ais:ais /etc/ais /var/lib/ais /var/log/ais
Create /etc/ais/ais.json, the initial cluster configuration. The file below was derived from the development deployment script (deploy/dev/local/aisnode_config.sh). Sections that are not listed use the AIS defaults; see the configuration documentation.
{
"features": "0",
"backend": {},
"timeout": {
"cplane_operation": "2s",
"max_keepalive": "5s",
"cold_get_conflict": "5s",
"max_host_busy": "20s",
"startup_time": "1m",
"join_startup_time": "3m",
"send_file_time": "5m",
"ec_streams_time": "10m",
"object_md": "2h"
},
"proxy": {
"primary_url": "http://127.0.0.1:8080",
"original_url": "http://127.0.0.1:8080",
"discovery_url": "",
"non_electable": false
},
"space": {
"cleanupwm": 65,
"lowwm": 75,
"highwm": 90,
"out_of_space": 95,
"batch_size": 32768,
"dont_cleanup_time": "120m"
},
"resilver": { "enabled": true },
"memsys": {
"min_free": "2gb",
"default_buf": "32kb",
"to_gc": "4gb",
"hk_time": "3m",
"min_pct_total": 0,
"min_pct_free": 0
},
"versioning": { "enabled": true, "validate_warm_get": false },
"net": {
"l4": { "proto": "tcp", "sndrcv_buf_size": -1 },
"http": {
"use_https": false,
"server_crt": "server.crt",
"server_key": "server.key",
"domain_tls": "",
"client_ca_tls": "",
"client_auth_tls": 0,
"idle_conn_time": "6s",
"idle_conns_per_host": 32,
"idle_conns": 256,
"write_buffer_size": 0,
"read_buffer_size": 0,
"chunked_transfer": true,
"skip_verify": false
},
"use_ipv6": false
},
"auth": {},
"log": {
"level": "3",
"max_size": "4mb",
"max_total": "128mb",
"flush_time": "10s",
"stats_time": "60s",
"to_stderr": false
}
}
Create the proxy's local configuration, /etc/ais/proxy.json. The public port must differ from the intra-cluster ports:
{
"confdir": "/var/lib/ais/proxy",
"log_dir": "/var/log/ais/proxy",
"host_net": {
"hostname": "127.0.0.1",
"port": "8080",
"port_intra_control": "9080",
"port_intra_data": "10080"
}
}
Create the target's local configuration, /etc/ais/target.json. The fspaths section lists the mountpaths, one per zoned XFS file system. The empty string is the optional mountpath label. Because the proxy runs on the same host in this example, the target uses different ports:
{
"confdir": "/var/lib/ais/target",
"log_dir": "/var/log/ais/target",
"host_net": {
"hostname": "127.0.0.1",
"port": "8081",
"port_intra_control": "9081",
"port_intra_data": "10081"
},
"fspaths": {
"/data/disk1": "",
"/data/disk2": "",
"/data/disk3": "",
"/data/disk4": ""
}
}
Step 4: Start the Cluster
Start the proxy first, then the target. -ntargets=1 tells the primary proxy
how many targets to wait for at startup. The test ran both processes in the
foreground in separate terminals; a systemd service setup was not tested.
$ aisnode -config=/etc/ais/ais.json -local_config=/etc/ais/proxy.json -role=proxy -ntargets=1
E 16:08:11.768505 config:3734 Warning: load initial global config from /etc/ais/ais.json
$ aisnode -config=/etc/ais/ais.json -local_config=/etc/ais/target.json -role=target
E 16:08:16.619206 config:3734 Warning: load initial global config from /etc/ais/ais.json
The warning appears on first start, when the node has no persisted cluster configuration yet. Point the ais CLI at the proxy and check the cluster:
$ export AIS_ENDPOINT=http://127.0.0.1:8080
$ ais show cluster
PROXY MEM USED(%) MEM AVAIL SYS CPU(%) UPTIME STATUS
p[gGwtZpiv][P] 0.03% 43.25GiB 0% - online
TARGET MEM USED(%) MEM AVAIL CAP USED(%) CAP AVAIL SYS CPU(%) UPTIME STATUS
t[uZyfRdPY] 0.03% 43.25GiB 0% 56.000GiB 9% - online
Cluster:
CLI Endpoint: http://127.0.0.1:8080
Proxies: 1
Targets: 1 (num disks: 4)
Capacity: used 24.00KiB (0%), available 56.00GiB
...
$ ais storage mountpath
uZyfRdPY
Used: min= 0%, avg= 0%, max= 0%
/data/disk1 /dev/sda(xfs)
/data/disk2 /dev/sdb(xfs)
/data/disk3 /dev/sdc(xfs)
/data/disk4 /dev/sdd(xfs)
The target log (/var/log/ais/target/aistarget.INFO) shows the disks found for each mountpath and the result of the extended attribute probe:
I 16:08:16.748688 dutils_linux:118 [/dev/sda]: [sda:{4096 0}]
...
W 16:08:16.754007 vinit:62 t[uZyfRdPY] VMD v1(uZyfRdPY, [/data/disk1 /data/disk4 /data/disk2 /data/disk3]) initialized
I 16:08:16.754068 fs:1302 volume: [56.00GiB, ssd]
I 16:08:16.781693 xattr_unix:102 max xattr size [ 65536 mp[/data/disk3, sdc, ssd] ]
AIS classifies each disk from its rotational attribute. The emulated zloop devices report ssd; SMR disks are expected to report hdd (not tested).
Step 5: Write and Read Data
Create a bucket, upload a 256 MiB object and eight 1 MiB objects, and list them with their checksums and locations:
$ ais bucket create ais://zbucket
"ais://zbucket" created
$ head -c 256M /dev/urandom > obj.bin
$ for i in $(seq 1 8); do head -c 1M /dev/urandom > small$i.bin; done
$ ais put obj.bin ais://zbucket/obj.bin
PUT "obj.bin" => ais://zbucket/obj.bin
$ ais put "small{1..8}.bin" ais://zbucket/small/ --yes
PUT 8 files => ais://zbucket/small/
$ ais ls ais://zbucket --props name,size,checksum,location
NAME SIZE CHECKSUM LOCATION
obj.bin 256.00MiB e19bec8e69d69270 t[uZyfRdPY]:mp[/data/disk1, sda, ssd]
small/small1.bin 1.00MiB 10cd336954e2e25f t[uZyfRdPY]:mp[/data/disk2, sdb, ssd]
small/small2.bin 1.00MiB 94dc556d0bbaaeaa t[uZyfRdPY]:mp[/data/disk4, sdd, ssd]
...
small/small8.bin 1.00MiB e0240cb5695941b4 t[uZyfRdPY]:mp[/data/disk3, sdc, ssd]
AIS computes an xxhash2 checksum on upload. Download the object, compare SHA-256 checksums with the original, and read it again with AIS's checksum validation:
$ ais get ais://zbucket/obj.bin obj.down
GET obj.bin from ais://zbucket as obj.down (256.00MiB)
$ sha256sum obj.bin obj.down
5843e6d8d2be6a22f25244f359c2b3280650fde36826a91244436352dd6f71d5 obj.bin
5843e6d8d2be6a22f25244f359c2b3280650fde36826a91244436352dd6f71d5 obj.down
$ ais get ais://zbucket/obj.bin /dev/null --checksum
GET and discard obj.bin from ais://zbucket (256.00MiB)
Step 6: Verify the Data Is on the Zoned Devices
Objects are regular files under <mountpath>/@ais/<bucket>/%ob/, with the
object name as the path:
$ cd /data/disk1
$ find . -maxdepth 4
.
./.ais.vmd
./.ais.smap
./@ais
./@ais/zbucket
./@ais/zbucket/%ch
./@ais/zbucket/%ut
./@ais/zbucket/%ds
./@ais/zbucket/%dw
./@ais/zbucket/%ob
./@ais/zbucket/%ob/obj.bin
./@ais/zbucket/%ob/small
./@ais/zbucket/%wk
./@ais/zbucket/%wk/small
./@ais/zbucket/%ec
./@ais/zbucket/%mt
./.ais.bmd.prev
./.ais.bmd
The object file has the realtime attribute (r). xfs_bmap shows its
extents on the realtime device; each realtime group (RG) is one zone. In
the test, the 256 MiB object filled one 256 MiB zone (RG 1) in 179 extents,
mostly 1 MiB each (xfs_bmap reports 512-byte units):
$ xfs_io -c lsattr @ais/zbucket/%ob/obj.bin
r---------------X @ais/zbucket/%ob/obj.bin
$ xfs_bmap -v @ais/zbucket/%ob/obj.bin
@ais/zbucket/%ob/obj.bin:
EXT: FILE-OFFSET RT-BLOCK-RANGE RG RG-OFFSET TOTAL
0: [0..2047]: 2621448..2623495 1 (8..2055) 2048
1: [2048..10239]: 2625544..2633735 1 (4104..12295) 8192
...
177: [523456..524279]: 3144904..3145727 1 (523464..524287) 824
178: [524280..524287]: 3145728..3145735 2 (0..7) 8
AIS stores the object metadata (checksum, version, size and copy locations) in the user.ais.lom extended attribute, 221 bytes for this object, and the target ID in user.ais.daemon_id on the mountpath root. XFS keeps extended attributes with its metadata, not on the realtime device:
$ getfattr -d -m - @ais/zbucket/%ob/obj.bin .
# file: @ais/zbucket/%ob/obj.bin
user.ais.lom=0sAgEr5QO6VgxF7QAAeHhoYXNoMuMvvQABZTE5YmVjOGU2OWQ2OTI3MOMvvQACMeMvvQADAAAAABAAAADjL70ABgAAAAAAAAAA4y+9AAcAA...
# file: .
user.ais.daemon_id="uZyfRdPY"
The zone report of the drive shows the full zone that holds the object and two open zones that hold the smaller files:
# blkzone report /dev/sda | sed -n 5,8p
start: 0x000200000, len 0x080000, cap 0x080000, wptr 0x000000 reset:0 non-seq:0, zcond: 1(em) [type: 2(SEQ_WRITE_REQUIRED)]
start: 0x000280000, len 0x080000, cap 0x080000, wptr 0x080000 reset:0 non-seq:0, zcond:14(fu) [type: 2(SEQ_WRITE_REQUIRED)]
start: 0x000300000, len 0x080000, cap 0x080000, wptr 0x001020 reset:0 non-seq:0, zcond: 2(oi) [type: 2(SEQ_WRITE_REQUIRED)]
start: 0x000380000, len 0x080000, cap 0x080000, wptr 0x001000 reset:0 non-seq:0, zcond: 2(oi) [type: 2(SEQ_WRITE_REQUIRED)]
Step 7: Mirror Objects Across Mountpaths
A bucket with mirroring enabled keeps the configured number of copies of each object on different mountpaths of the same target. Enabling it starts a job that copies the existing objects:
$ ais bucket props set ais://zbucket mirror.enabled=true mirror.copies=2
"mirror.enabled" set to: "true" (was: "false")
Bucket props successfully updated.
Started job n41CQ4qF1vX. To monitor the progress, run 'ais show job n41CQ4qF1vX'
$ ais object show ais://zbucket/obj.bin --props copies,location
PROPERTY VALUE
copies 2 [/data/disk4 /data/disk1]
location t[uZyfRdPY]:mp[/data/disk1, sda, ssd]
$ xfs_io -c lsattr /data/disk4/@ais/zbucket/%ob/obj.bin
r---------------X /data/disk4/@ais/zbucket/%ob/obj.bin
Erasure coding places slices on different targets and needs more than one target. With a single target, the request fails:
$ ais bucket props set ais://zbucket ec.enabled=true ec.data_slices=2 ec.parity_slices=1
Error: ErrNotEnoughTargets: not enough target nodes: EC configuration (D = 2, P = 1) requires at least 4 targets (have 1).
Erasure coding was not tested on zoned XFS for this guide.
AIStore Disk Checks on Zoned XFS
The tested target accepted zoned XFS mountpaths without changes:
-
File system type and ID. At startup and on
mountpath attach, AIS records the device, file system type and file system ID (statfs f_fsid) of each mountpath and refuses two mountpaths on the same file system. Each zoned XFS file system reports its own ID:$ ais storage mountpath show --json..."/data/disk1": {"mountpath_label": "","fs": {"Fs": "/dev/sda","FsType": "xfs","FsID": "66339,0"},"disks": ["sda"],...Attaching a bind mount of an already-used file system was rejected:
$ ais storage mountpath attach uZyfRdPY=/data/disk1xError: FsID 66340,0: filesystem sharing is not allowed: mp[/data/disk1x, fs=/dev/sdb/xfs, unknown] vs "/data/disk2" -
Extended attributes. The target probes the maximum xattr size on startup. Zoned XFS reported 64 KiB (
max xattr size [ 65536 ...]in the log above). -
Direct I/O. The filesystem health checker (FSHC) reads existing files and writes temporary files in
<mountpath>/.$deleted/fshc-on-err/with O_DIRECT, followed by fsync. Run it on demand per mountpath:$ ais storage mountpath fshc uZyfRdPY=/data/disk1t[uZyfRdPY]: started filesystem health check on mountpath "/data/disk1"The target log shows the result:
I 16:09:22.139013 fshc:301 3. write 4 temp files to /data/disk1/.$deleted/fshc-on-errI 16:09:22.187521 fshc:329 4. remove /data/disk1/.$deleted/fshc-on-errI 16:09:22.187559 fshc:185 mp[/data/disk1, sda, ssd] appears to be healthyAll four mountpaths passed. Tracing the target with strace confirmed
openat(..., O_RDWR|O_CREAT|O_TRUNC|O_DIRECT|O_CLOEXEC)calls for the health-check files. -
No preallocation or reflink. AIS does not call fallocate() or use reflink copies; no fallocate calls were seen while tracing uploads and health checks. Uploads write a work file under
%wk/with buffered I/O. The zoned XFS limitations on preallocation, reflink and quotas do not affect it.
Production Notes
-
One file system per drive. Format each SMR disk or ZNS namespace as a separate zoned XFS file system and list each as a mountpath. Do not combine zoned drives with LVM or RAID. Use mirroring within a target, or erasure coding or mirroring across targets, for redundancy.
-
Metadata on conventional storage. Keep the proxy and target confdir directories on an SSD. The target also writes small copies of the cluster map and bucket metadata (.ais.smap, .ais.bmd, .ais.vmd) to each mountpath root; these files land on the zoned device, and their XFS metadata and the object xattrs stay on the conventional part of the file system.
-
Uneven fill across mountpaths. AIS selects the mountpath by object name. In a test with about 700 objects of 64 MiB, one mountpath reached 93% usage while the others were at 78% to 80%. Usage close to full increases the work XFS garbage collection has to do on that drive. The AIS capacity watermarks (space.cleanupwm, lowwm, highwm, out_of_space, 65/75/90/95% by default) apply to the target as a whole: above highwm, the target logs
_not_ running store cleanup: ... high-wmwarnings. -
Space reclaim after deletes. Deleted objects are removed from the file system immediately. In the test,
dfusage dropped within seconds afterais object rmandais rmo --prefix. Deleting large objects left whole zones without valid data, and XFS reset them; an XFS trace of a refill after deleting about 20 GiB showed 16 to 20 zone resets per drive and no garbage-collection victim selection. Workloads that delete small objects mixed with long-lived ones leave partially valid zones that XFS garbage collection must copy. Raising zonegc_low_space starts garbage collection earlier:# echo 5 > /sys/fs/xfs/sda/zoned/zonegc_low_space -
Open zones. By default, XFS writes to at most the device's open-zone limit minus one (14 for the test devices, see
/sys/fs/xfs/sda/zoned/max_open_zones). Concurrent uploads to one mountpath share this limit. For SMR disks, a lower max_open_zones mount option can improve throughput. See XFS tuning. The other XFS mount options from the AIS performance guide (largeio, swalloc, allocsize) were not tested on zoned XFS. -
Restarts. After a cluster shutdown (
ais cluster shutdown), unmount and remount, the target found its mountpaths and all objects. A downloaded object matched its original SHA-256 checksum. -
Tested load. Repeated overwrites of 40 objects of 64 MiB, a fill to 85% with interleaved objects, deletion of half of them and a refill to 93% completed without errors. All 695 remaining objects matched their SHA-256 checksums.
Troubleshooting
-
"go: download go1.27 for linux/amd64: toolchain not available". GOTOOLCHAIN=auto tries to download the version in go.mod (go 1.27) without a patch level. Name a released toolchain, such as
GOTOOLCHAIN=go1.27.3. -
"http error code 'Service Unavailable'" right after startup. The cluster is still starting (
ais show clustershows the cluster-not-started-yet alert). Wait a few seconds and retry. -
"filesystem sharing is not allowed". Two mountpaths are on the same file system. Use one zoned XFS file system per mountpath.
-
ais putwith several files asks "[Y/N]" and cancels. Multi-file uploads ask for confirmation. Add--yesin scripts. -
Object files not on the zoned device. If xfs_io -c lsattr does not show the
rflag for files under%ob/, check that mkfs.xfs reported zoned=1 and that the mountpath directory has the rtinherit flag (t):$ xfs_io -c lsattr /data/disk1-------t--------X /data/disk1