Red Hat Training

A Red Hat training course is available for Red Hat Ceph Storage

Chapter 6. Known Issues

This section documents known issues found in this release of Red Hat Ceph Storage.

The Calamari API sometimes incorrectly prints standard output as an error

When using the Calamari /cli endpoint, the Calamari API response swaps the standard output (stdout) and standard error (stderr). As a consequence, the err field contains information from the stdout field. This behavior does not have any impact on the Calamari API commands. (BZ#1287904, BZ#1287905)

Renaming snapshots returns errors on overloaded clusters

When a Ceph storage cluster is overloaded and an image is currently in-use for I/O, the cluster might not be able to service rename requests in a timely fashion. Consequently, the RADOS Block Device (RBD) CLI keeps sending the rename request every 5 seconds because it receives a message that the request timed out. This behavior causes error log messages to be returned within the logs of the process performing I/O on the image. To avoid this problem, do not rename snapshots when the cluster is overloaded. (BZ#1340772)

Multi-site configuration of the Ceph Object Gateway sometimes fails when options are changed at runtime

When the rgw md log max shards and rgw data log num shards options are changed at runtime in multi-site configuration of the Ceph Object Gateway, the radosgw process terminates unexpectedly with a segmentation fault.

To avoid this issue, do not change the aforementioned options at runtime, but set them during the initial configuration of the Ceph Object Gateway. (BZ#1330952)

Calamari hangs when interactive commands are issued

The Calamari REST API incorrectly includes the following interactive commands:

  • rbd import
  • rbd import-diff
  • rbd journal import
  • rbd merge-diff
  • rbd watch

When these commands are executed from Calamari, Calamari becomes unresponsive because it waits for an action from the user.

To work around this issue, do not execute the aforementioned commands from the Calamari REST API and use the command line instead. (BZ#1354459)

Bucket creation fails after upgrading Red Hat Ceph Storage 1.3 to 2.0

After upgrading an Object Gateway node from Red Hat Ceph Storage 1.3 to 2.0, an attempt to create a bucket fails. To work around this issue, restart the radosgw service and create the bucket again. (BZ#1352888)

Unable to write data on a promoted image

In RBD mirroring configuration, an image can be demoted to non-primary on the local cluster and promoted to primary on the remote cluster. If this happens and the rbd-mirror daemon is not restarted on the remote cluster, it is not possible to write data on the promoted image because rbd-daemon considers the demoted image on the local cluster to be the primary one.

To avoid this issue, restart the rbd-mirror daemon to gain the read/write access to the promoted image. (BZ#1365648)

Mirroring image metadata is not supported

Image metadata are not currently replicated to a peer cluster. (BZ#1344212)

The radosgw-admin command sometimes returns unnecessary error messages

The log level of certain common error messages that are returned by the radosgw-admin command is set to 0. Consequently, these messages appears even though radosgw-admin command succeeds. These messages can be safely ignored. (BZ#1364353)

Ansible fails to install OSDs if they point to directories

Ansible does not support installation of OSDs that point to directories and not to partitions. As a consequence, an attempt to install such OSDs fails. (BZ#1361228)

The default value of journal_size is 0

By default, the journal_size option in the ceph-ansible utility is set to 0. This option is mandatory and Red Hat strongly recommends to reserve enough disk space for the journal, at least 5 GB depending on the size of your cluster. Using a journal smaller than 5 GB leads to critical performance issues and production stability problems.

For more details, see the Journal Settings section in the Configuration Guide for Red Hat Ceph Storage 2. (BZ#1359889)

Certain features available in ceph-ansible are not supported

The ceph-ansible utility contains the following features that are not supported with Red Hat Ceph Storage 2.0:

  • the purge-cluster feature
  • the restapis role (BZ#1366394)

Sync point snapshots are not properly removed after a failover or failback

In a RADOS Block Device (RBD) mirroring configuration, it is possible that sync point snapshots created by the rbd-mirror daemon incorrectly remain on the clusters after a failover or failback. These snapshots can be safely removed if an image synchronization is not in progress. (BZ#1350003)

Buckets sometimes have incorrect time stamps

Buckets created by the Simple Storage Service (S3) API on the Ceph Object Gateway before mounting the Ganesha NFS interface have incorrect time stamps. (BZ#1359404)

The UID and GID of NFS directories is 4294967294 when ID mapping is not properly configured

When ID mapping between a client node and the node with the NFS interface for the Ceph Object Gateway is not properly configured, the UID and GID of newly created directories are set to 4294967294, which is the default. This behavior is equivalent to the all_squash option of the knfsd NFS server. (BZ#1359407)

Ansible fails to add a monitor to an upgraded cluster

An attempt to add a monitor to a cluster by using the Ansible automation application after upgrading the cluster from Red Hat Ceph Storage 1.3 to 2.0 fails on the following task: 

TASK: [ceph-mon | collect admin and bootstrap keys]

This happens because the original monitor keyring was created with the mds "allow" capability while the newly added monitor requires a keyring with the mds "allow *" capability.

To work around this issue, after installing the ceph-mon package, manually copy the administration keyring from an already existing monitor node to the new monitor node: 

scp /etc/ceph/<cluster_name>.client.admin.keyring <target_host_name>:/etc/ceph

For example: 

# scp /etc/ceph/ceph.client.admin.keyring node4:/etc/ceph

Then use Ansible to add the monitor as described in the Adding a Monitor with Ansible section of the Administration Guide for Red Hat Ceph Storage 2. (BZ#1357292)

Certain maintenance image operations are incorrectly allowed on non-primary images

With RADOS Block Device (RBD) mirroring enabled, non-primary images are expected to be read-only. Under certain conditions, the rbd command does not properly restrict rbd maintenance operations against non-primary images. The affected operations include:

  • updating snapshots
  • resizing images
  • renaming and creating clones using the non-primary image as the parent
  • disabling image features

These operations should be disallowed on non-primary images. To avoid this issue, make sure to perform the maintenance operations only against the primary image instance. (BZ#1348928, BZ#1352878, BZ#1353877, BZ#1349332)

The NFS interface for the Ceph Object Gateway does not show bucket size or number of blocks

The NFS interface of the Ceph Object Gateway lists buckets as directories. However, the interface always shows that the directory size and the number of blocks is 0, even if some data is written to the buckets. (BZ#1359408)

Certain image features are not supported with the RBD kernel module

The following image features are not supported with the current version of the RADOS Block Device (RBD) kernel module (krbd) that is shipped with the Red Hat Enterprise Linux 7.2:

  • object-map
  • deep-flatten
  • journaling
  • exclusive-lock
  • fast-diff

However, by default the ceph-installer utility creates RBDs with the aforementioned features enabled. As a consequence, an attempt to map the kernel RBDs by running the rbd map command fails.

To work around this issue, disable the unsupported features by setting the rbd_default_features option in the Ceph configuration file for kernel RBDs or dynamically disable them by running the following command: 

rbd feature disable <image> <feature>

This issue is a limitation only in kernel RBDs, and the features work as expected with user-space RBDs. (BZ#1340080)

The rbd bench write command fails when --io-size is greater than or equal to the image size

The rbd bench-write --io-size <size> <image> command fails with a segmentation fault when the size specified by the --io-size option is greater than or equal to the image size.

To avoid this issue, make sure that the value of --io-size is smaller than the image size. (BZ#1362014)

FUSE clients cannot be mounted permanently on Red Hat Enterprise Linux 7.2

The util-linux package shipped with Red Hat Enterprise Linux 7.2 does not support mounting CephFS Filesystem in Userspace (FUSE) clients in the /etc/fstab file. Red Hat Enterprise Linux 7.3 will include a new version of util-linux that will support mounting CephFS FUSE clients permanently. (BZ#1360849)

Image synchronization starts from the beginning after restarting rbd-mirror

When the rbd-mirror daemon is restarted during image synchronization, the synchronization starts from the beginning.

To avoid this issue, do not restart rbd-mirror during image synchronization. (BZ#1348940)

Results form deep scrubbing are overwritten by shallow scrubbing

When performing shallow scrubbing after deep scrubbing, results from deep scrubbing are overwritten by results from shallow scrubbing. As a consequence, the deep scrubbing results are lost. There is no workaround for this issue yet. (BZ#1330023)

Calamari sometimes does not respond when sending a PATCH Request

The Calamari API does not respond when making PATCH requests to /api/v2/cluster/FSID/osd/OSD_ID if the requests does not change any fields on the OSD from their present values. (BZ#1338688)

RBD mirroring is not disabled if peer clusters are not removed first

Disabling of RADOS Block Device (RBD) mirroring in pool mode consists of two steps:

  1. Removing peer clusters by using the rbd mirror pool remove command.
  2. Disabling pool mirroring by using the rbd mirror pool disable command.

If pool mirroring is disabled before removing the peer clusters, all the existing images on the mirrored pool stop being replicated but the pool remains mirrored. As a consequence, newly created images on the pool will be still replicated.

In addition, an error message similar to the following one is returned: 

2016-08-02 10:30:31.823726 7f96fd7e8d80 -1 librbd: Failed to set mirror mode: (16) Device or resource busy

To avoid this issue, first remove peer clusters and then disable pool mirroring. (BZ#1362647)

An LDAP user can access buckets created by a local RGW user with the same name

The RADOS Object Gateway (RGW) does not differentiate between a local RGW user and an LDAP user with the same name. As a consequence, the LDAP user can access the buckets created by the local RGW user.

To work around this issue, use different names for RGW and LDAP users. (BZ#1361754)

POSIX ACL support is disabled by default in CephFS FUSE clients

Support for Access Control Lists (ACL) is disabled by default for Ceph File Systems (CephFS) mounted as FUSE clients. To use the ACL feature with FUSE clients, enable it manually. For details, see the Limitations section in the Ceph File System Guide for Red Hat Ceph Storage 2.

In addition, ACL in CephFS kernel clients is supported on Red Hat Enterprise Linux with kernel version kernel-3.10.0-327.18.2.el7 or later. (BZ#1342751)

Failed instances of the ceph-disk service

After installation of a new Ceph storage cluster, failed instances of the ceph-disk service appear. This is because the service is started twice - once to activate the data partition and once to activate the journal partition. After the disk activation, one of these instances fails because of limited resources. This is an expected behavior and the failed instance does not have any impact on the disk activation. (BZ#1326740)

The Calamari API returns the "server error (500)" error when changing the take step

When changing a CRUSH rule, modifying the take step type to any other value than take causes the Calamari API to return the "server error (500)" error.

To avoid this issue, do not change the take step to any other value. (BZ#1329216)

Client installation by using Ansible is not supported

The current version of the ceph-ansible utility does not support installation of the Ceph command-line interface (CLI). To use the Ceph CLI, install it manually. For details, see the Client Installation chapter in in the Installation Guide for Red Hat Enterprise Linux or the Client Installation chapter in in the Installation Guide for Ubuntu. (BZ#1335308)

S3 versioning cannot be set on buckets when accessing a non-master zone

Amazon Simple Storage Service (S3) object versioning cannot be set on buckets when accessing any zone other than the metadata master zone. (BZ#1350522)

An error message is returned when downloading an S3 multipart file by using Swift

When uploading a multiple part file by using the Simple Storage Service (S3) and then downloading it by using the Swift service, the following error message is returned: 

Error downloading object: md5sum != etag

Despite the message, the upload and download operations succeed, and the message can be safely ignored. (BZ#1361044)

The rados list-inconsistent-obj command does not highlight inconsistent shards when it could have

The output of the rados list-inconsistent-obj command does not explicitly show which shard is inconsistent when it could have. (BZ#1363949)

Restart of the radosgw service on clients is needed after rebooting the cluster

After rebooting the Ceph cluster, the radosgw service must be restarted on the Ceph Object Gateway clients to restore the connection with the cluster. (BZ#1363689)

Ansible does not support adding encrypted OSDs

The current version of the ceph-ansible utility does not support adding encrypted OSD nodes. As a consequence, an attempt to perform asynchronous updates between releases by using the rolling_update playbook fails to upgrade encrypted OSD nodes. In addition, Ansible returns the following error message during the disk activation task: 

mount: unknown filesystem type 'crypto_LUKS'

To work around this issue, open the rolling_update.yml file located in the Ansible working directory and find all instances of the roles: list. Then remove or comment out all roles (ceph-mon, ceph-rgw, ceph-osd or, ceph-mds) except the ceph-common role from the lists, for example: 

roles:
    - ceph-common
    #- ceph-mon
Note

Make sure to edit all instances of the roles: list in the rolling_update.yml file.

Then run the rolling_update playbook to update the nodes. (BZ#1366808)

Dynamic feature updates are not replicated

When a feature is disabled or enabled on an already existing image and the image is mirrored to a peer cluster, the feature is not disabled or enabled on the replicated image. (BZ#1344262)

Users created by using the Calamari API do not have permissions to run the API commands

When a user is created by using the Calamari REST API (api/v2/user), the user does not have permissions to run most of the Calamari API commands. Consequently, an attempt to run the commands fails with the following error message: 

"You do not have permission to perform this action"

To work around this issue, use the calamari-ctl add_user command from the command line when creating new users. (BZ#1356872)

Ansible does not support removing monitor or OSD nodes

The current version of the ceph-ansible utility does not support removing monitor or OSD nodes. To remove monitor or OSD nodes from a cluster, use the manual procedure. For more information, see the Administration Guide for Red Hat Ceph Storage 2. (BZ#1335569)

Error messages are returned after disabling journaling

When the journaling feature is disabled on an image that was previously mirrored by the rbd-mirror daemon, an error message similar to the following is returned on the primary cluster: 

2016-06-15 22:10:40.462481 7fed3d10b700 -1 rbd::mirror::ImageReplayer: 0x7fecd8003b80 [1/29d86f79-7bba-4316-9ab9-c8a3f600e0f2] operator(): start failed: (2) No such file or directory

These harmless error messages indicates that the image journal was deleted because journaling was disabled and can be safely ignored. (BZ#1346946)

Calamari sometimes incorrectly outputs "null" as a value

When the Calamari REST-based API is used to get details of a CRUSH rule in the Ceph cluster, the output contains "null" as a value for certain fields in the steps section of the CRUSH rule. The fields containing null values can be safely ignored for the respective steps in the CRUSH rule. However, do not use "null" as a value for any field when doing a PATCH operation. Using null values in such a case causes the operation to fail. (BZ#1342504)

The Ceph Object Gateway is restarted after switching the zone from master to non-master

When a non-master zone is promoted to the master zone, all I/0 requests become unresponsive until the radosgw process is restarted on both zones. Consequently, the I/0 requests time out. (BZ#1359712)

Images are synchronizing from the beginning after their demotion and promotion

With RADOS Block Device (RBD) mirroring enabled, an image can be demoted to non-primary on one cluster and promoted to primary on a peer cluster. When this happens, the rbd-mirror daemon starts to synchronize the newly demoted image with the newly promoted image even though the image has been already successfully synchronized before the demotion and promotion of the image. This behavior is not optimal and will be fixed in a future release. (BZ#1349955)

Ansible does not properly handle unresponsive tasks

Certain tasks, for example adding monitors with the same host name, cause the ceph-ansible utility to become unresponsive. Currently, there is no timeout set after which the unresponsive tasks is marked as failed. (BZ#1313935)

The Calamari REST-based API fails to edit user details

An attempt to use the Calamari REST-based API to edit user details fails with an error. To change user details, use the calamari-ctl command-line utility. (BZ#1338649)

Object sync requests are sometimes skipped

In multi-site configurations of the Ceph Object Gateway, a non-master zone can be promoted to the master zone. In most cases, the master zone’s gateway or gateways are still running when this happens. However, if the gateways are down, it can take up to 30 seconds after their restart until the gateways notice that another zone was promoted. During this time, the gateways can miss changes to buckets that occur on other zones. Consequently, object sync requests are skipped.

To work around this issue, pull the new master’s period to the old master zone before restarting the old master zone: 

$ radosgw-admin period pull --remote=<new-master-zone-id>

For details on pulling the period, see the Ceph Object Gateway Guide for Red Hat Enterprise Linux or the Ceph Object Gateway Guide for Ubuntu. (BZ#1362639)