** Description changed: [Impact] - * If some commands like "iscsiadm -m session" fail, the thread can abort immediately without updating any counters like failed_logins or stopped_threads properly, because there are no try-except block to catch exceptions. - * The main thread keeps waiting until these counters are updated, and this results in stuck of volume attachment process. + * If some commands like "iscsiadm -m session" fail, the thread can abort immediately without updating any counters like failed_logins or stopped_threads properly, because there are no try-except block to catch exceptions. + * The main thread keeps waiting until these counters are updated, and this results in stuck of volume attachment process. [Test Case] - * Deploy Cinder with an iSCSI driver and multipath enabled. - - * TBC. + * Deploy Cinder with a backend that uses an iSCSI driver and configure Multipath + * Attach a volume to an instance (first attachment for a period + * See log line like: + 2021-12-01 00:23:24.044 2679 WARNING os_brick.initiator.connectors.iscsi [...] iscsiadm stderr output when getting sessions: iscsiadm: No active sessions. + * Volume attachment never completes + * Passing test: Log line appears but volume attachment succeeds. [Where problems could occur] - * Change primarily introduces error handling and doesn't change implementation details. - As such we may see an error condition logged. + * Change primarily introduces error handling and doesn't change implementation details. + As such we may see an error condition logged. [Other Info] - * - + * - --- Original Description --- Currently we execute login to iscsi portals and device discovery in multiple threads concurrently when multipath is enabled. However if some commands like "iscsiadm -m session" fail, the thread can abort immediately without updating any counters like failed_logins or stopped_threads properly, because there are no try-except block to catch exceptions. However the main thread keeps waiting until these counters are updated, and this results in stuck of volume attachment process. This issue was initially reported in downstream bug https://bugzilla.redhat.com/show_bug.cgi?id=1923975 , and maybe is caused by a bug in iscsiadm command. However we should handle the error more properly because current behavior requires operators to restart services like cinder-volume to resolve the stuck.
** Description changed: [Impact] * If some commands like "iscsiadm -m session" fail, the thread can abort immediately without updating any counters like failed_logins or stopped_threads properly, because there are no try-except block to catch exceptions. * The main thread keeps waiting until these counters are updated, and this results in stuck of volume attachment process. [Test Case] * Deploy Cinder with a backend that uses an iSCSI driver and configure Multipath - * Attach a volume to an instance (first attachment for a period - * See log line like: + * Attach a volume to an instance (first attachment for a period + * See log line like: 2021-12-01 00:23:24.044 2679 WARNING os_brick.initiator.connectors.iscsi [...] iscsiadm stderr output when getting sessions: iscsiadm: No active sessions. - * Volume attachment never completes - * Passing test: Log line appears but volume attachment succeeds. + * Volume attachment never completes + * Passing test: Log line appears but volume attachment succeeds. [Where problems could occur] * Change primarily introduces error handling and doesn't change implementation details. As such we may see an error condition logged. - - [Other Info] - - * - --- Original Description --- Currently we execute login to iscsi portals and device discovery in multiple threads concurrently when multipath is enabled. However if some commands like "iscsiadm -m session" fail, the thread can abort immediately without updating any counters like failed_logins or stopped_threads properly, because there are no try-except block to catch exceptions. However the main thread keeps waiting until these counters are updated, and this results in stuck of volume attachment process. This issue was initially reported in downstream bug https://bugzilla.redhat.com/show_bug.cgi?id=1923975 , and maybe is caused by a bug in iscsiadm command. However we should handle the error more properly because current behavior requires operators to restart services like cinder-volume to resolve the stuck. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1915678 Title: [SRU] iSCSI+Multipath: Volume attachment hungs if sessiong scanning fails To manage notifications about this bug go to: https://bugs.launchpad.net/cloud-archive/+bug/1915678/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
