This fixes the following:
- Unchecked thread growth in RemoteEndHostEndPoint
- Potential NPE while finding EP for a storage/scope
Unbounded thread growth can be reproduced with following findings:
- Every unreachable template would produce 6 new threads (in a single
ScheduledExecutorService instance) spaced by 10 seconds
- Every reachable template url without the template would produce 1 new
thread (and one ScheduledExecutorService instance), it errors out quickly without
causing more thread growth.
- Every valid url will produce upto 10 threads as the same ep (endpoint
instance) will be reused to query upload/download (async callback)
progresses.
Every RemoteHostEndPoint instances creates its own
ScheduledExecutorService instance which is why in the jstack dump, we
see several threads that share the prefix RemoteHostEndPoint-{1..10}
(given poolsize is defined as 10, it uses suffixes 1-10).
This fixes the discovered thread leakage with following notes:
- Instead of ScheduledExecutorService instance, a cached pool could be
used instead and was implemented, and with `static` scope to be reused
among other future RemoteHostEndPoint instances.
- It was not clear why we would want to wait when we've Answers returned
from the remote EP, and therefore a scheduled/delayed Runnable was
not required at all for processing answers. ScheduledExecutorService
was therefore not really required, moved to ExecutorService instead.
- Another benefit of using a cached pool is that it will shutdown
threads if they are not used in 60 seconds, and they get re-used for
future runnable submissions.
- Caveat: the executor service is still unbounded, however, the use-case
that this method is used for short jobs to check upload/download
progresses fits the case here.
- Refactored CmdRunner to not use/reference objects from parent class.
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
- Migrate to embedded Jetty server.
- Improve ServerDaemon implementation.
- Introduce a new server.properties file for easier configuration.
- Have a single /etc/default/cloudstack-management to configure env.
- Reduce shaded jar file, removing unnecessary dependencies.
- Upgrade to Spring 5.x, upgrade several jar dependencies.
- Does not shade and include mysql-connector, used from classpath instead.
- Upgrade and use bountcastle as a separate un-shaded jar dependency.
- Remove tomcat related configuration and files.
- Have both embedded UI assets in uber jar and separate webapp directory.
- Refactor systemd and init scripts, cleanup packaging.
- Made cloudstack-setup-databases faster, using `urandom`.
- Remove unmaintained distro packagings.
- Moves creation and usage of server keystore in CA manager, this
deprecates the need to create/store cloud.jks in conf folder and
the db.cloud.keyStorePassphrase in db.properties file. This also
remove the need of the --keystore-passphrase in the
cloudstack-setup-encryption script.
- GZip contents dynamically in embedded Jetty
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
This implements a CloudStack Prometheus exporter as a plugin, that serves
metrics on a HTTP port.
New global settings:
1. prometheus.exporter.enable - (default: false), Enable the prometheus
exporter plugin, management server restart needed.
2. prometheus.exporter.port - (default: 9595), The prometheus exporter
server port.
3. prometheus.exporter.allowed.ips - (default: 127.0.0.1), List of comma
separated prometheus server ips (with no spaces) that should be allowed to
access the URLs.
The following list of metrics are provided per pop (zone) with the exporter:
• Per host:
o CPU cores: used, total
o CPU usage: used, total (in MHz)
o Memory usage: used, total (in MiBs)
o Total VMs running on the host
• CPU cores: allocated (per zone)
• CPU usage: allocated (per zone, in MHz)
• Memory usage: allocated (per zone, in MiBs)
• Hosts: online, offline, total
• VMs: in all states -- starting, running, stopping, stopped, destroyed,
expunging, migrating, error, unknown
• Volumes: ready, destroyed, total
• Primary Storage Pool: (Disk size) used, allocated, unallocated, total (in GiBs)
• Secondary Storage Pool: (Disk size) used, allocated, unallocated, total (in GiBs)
• Private IPs: allocated, total
• Public IPs: allocated, total
• Shared Network IPs: allocated, total
• VLANs: allocated, total
Additional metrics for the environment:
• Summed domain (level=1) limit for CPU cores
• Summed domain (level=1) limit for memory/ram
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
Snapshots are not deleted resulting unexpected storage consumption in case of VMware.
Steps to reproduce this issue :
In VMware setup, create a snapshot of volume say Snap1.
After successful creation of snapshot Snap1, create new snapshot of same volume say Snap2.snapshots
While Snap2 is in BackingUp state, delete Snap1.
Snap1 will disappear from Web UI, but when we check secondary storage, files associated with Snap1 still persists even after cleanup job is performed.
In snapshot_store_ref table in DB, Snap1 will be in ready state instead of Destroyed.
Also, in snapshots table, status of Snap1 will be Destroyed but removed column will be null and will never change to the date of snapshot removal.
Fix for this issue :
In VMware, snapshot chain is not maintained, instead full snapshot is taken every time.
So, it makes sense not to assign parent snapshot id for the snapshot. In this way, every snapshot will be individual and can be deleted successfully whenever required.
This fixes the agreed upon url on download.cloudstack.org in various
sql files and misc scripts.
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
CLOUDSTACK-9660: NPE while destroying volumes during 1000 VMs deploy and destroy tests
NPE is seen as VM destroy and storage cleanup threads try to remove the same root volume. Fix is to handle
only non-root volumes in storage cleanup thread, root volumes will be handled as part of VM destroy.
* pr/1825:
CLOUDSTACK-9660: NPE while destroying volumes during 1000 VMs deploy and destroy tests NPE is seen as VM destroy and storage cleanup threads try to remove the same root volume. Fix is to handle only non-root volumes in storage cleanup thread, root volumes will be handled as part of VM destroy.
Signed-off-by: Rajani Karuturi <rajani.karuturi@accelerite.com>
This closes#1644
* 4.9:
CLOUDSTACK-4858 Honors the snapshot.backup.rightafter configuration variable Unhides snapshot.backup.rightafter from global configuration
CLOUDSTACK-4858 Honors the snapshot.backup.rightafter configuration variable
Unhides snapshot.backup.rightafter from global configuration
If snapshot.backup.rightafter is set to false (defaults to true), snapshots are
not backed up to secondary storage.
This is the same as PR #1644 applied to 4.9, as per @jburwell
* pr/1697:
CLOUDSTACK-4858 Honors the snapshot.backup.rightafter configuration variable Unhides snapshot.backup.rightafter from global configuration
Signed-off-by: Rajani Karuturi <rajani.karuturi@accelerite.com>
[4.10] CLOUDSTACK-8746: VM Snapshotting implementation for KVM
* pr/977:
Fixes for testing VM Snapshots on KVM. Related to PR 977
CLOUDSTACK-8746: vm snapshot implementation for KVM
Signed-off-by: Rajani Karuturi <rajani.karuturi@accelerite.com>
CLOUDSTACK-9619: Updates for SAN-assisted snapshotsThis PR is to address a few issues in #1600 (which was recently merged to master for 4.10).
In StorageSystemDataMotionStrategy.performCopyOfVdi we call getSnapshotDetails. In one such scenario, the source snapshot in question is coming from secondary storage (when we are creating a new volume on managed storage from a snapshot of ours thats on secondary storage).
This usually worked in the regression tests due to a bit of "luck": We retrieve the ID of the snapshot (which is on secondary storage) and then try to pull out its StorageVO object (which is for primary storage). If you happen to have a primary storage that matches the ID (which is the ID of a secondary storage), then getSnapshotDetails populates its Map<String, String> with inapplicable data (that is later ignored) and you dont easily see a problem. However, if you dont have a primary storage that matches that ID (which I didnt today because I had removed that primary storage), then a NullPointerException is thrown.
I have fixed that issue by skipping getSnapshotDetails if the source is coming from secondary storage.
While fixing that, I noticed a couple more problems:
1) We can invoke grantAccess on a snapshot thats actually on secondary storage (this doesnt amount to much because the VolumeServiceImpl ignores the call when its not for a primary-storage driver).
2) We can invoke revokeAccess on a snapshot thats actually on secondary storage (this doesnt amount to much because the VolumeServiceImpl ignores the call when its not for a primary-storage driver).
I have corrected those issues, as well.
I then came across one more problem:
When using a SAN snapshot and copying it to secondary storage or creating a new managed-storage volume from a snapshot of ours on secondary storage, we attach to the SR in the XenServer code, but detach from it in the StorageSystemDataMotionStrategy code (by sending a message to the XenServer code to perform an SR detach). Since we know to detach from the SR after the copy is done, we should detach from the SR in the XenServer code (without that code having to be explicitly called from outside of the XenServer logic).
I went ahead and changed that, as well.
JIRA Ticket:
https://issues.apache.org/jira/browse/CLOUDSTACK-9619
* pr/1749:
CLOUDSTACK-9619: Updates for SAN-assisted snapshots
Signed-off-by: Rajani Karuturi <rajani.karuturi@accelerite.com>
(1) add support to create/delete/revert vm snapshots on running vms with QCOW2 format
(2) add new API to create volume snapshot from vm snapshot
(3) delete metadata of vm snapshots before stopping/migrating and recover vm snapshots after starting/migrating
(4) enable deleting of VM snapshot on stopped vm or vm snapshot is not listed in qcow2 image.
(5) enable smoke tests for vmsnaphsots on KVM
store ref table so builtin template is never downloaded completely
In handleSysTemplateDownload method creating template only if there exists no entry
handleTemplateSync will take care of other scenario
- Bump spring-framework version to 4.x and Jetty to version that runs with JDK8
- Bump servet dependency version
- Migrate spring xmls to version 4, fixes schema locations that are 3.0
dependent in various xmls.
- Fix failing tests due to spring upgrade
(Thanks @marcaurele Marc-Aurèle Brothier for fixing them)
* Fix test DeploymentPlanningManagerImplTest
* Fix GloboDNS test
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
NPE is seen as VM destroy and storage cleanup threads try to remove the same root volume. Fix is to handle
only non-root volumes in storage cleanup thread, root volumes will be handled as part of VM destroy.
CLOUDSTACK-9627 Fix template sync for region store.When using a region store like Swift or S3 as secondary storage,
the `zoneId` can be null. This causes an exception when we try
to convert it to a `long`. This fix guards against that.
Before this fix, if you restart the management server, all the templates
would change to "NOT READY" because the code which syncs the NFS cache
and the object store crashes due to the above mentioned issue.
This PR fixes that.
* pr/1772:
CLOUDSTACK-9627:Fix template sync for region store
Signed-off-by: Rohit Yadav <rohit.yadav@shapeblue.com>
This PR adds an ability to Pass a new parameter, locationType,
to the “createSnapshot” API command. Depending on the locationType,
we decide where the snapshot should go in case of managed storage.
There are two possible values for the locationType param
1) `Standard`: The standard operation for managed storage is to
keep the snapshot on the device. For non-managed storage, this will
be to upload it to secondary storage. This option will be the
default.
2) `Archive`: Applicable only to managed storage. This will
keep the snapshot on the secondary storage. For non-managed
storage, this will result in an error.
The reason for implementing this feature is to avoid a single
point of failure for primary storage. Right now in case of managed
storage, if the primary storage goes down, there is no easy way
to recover data as all snapshots are also stored on the primary.
This features allows us to mitigate that risk.
Unhides snapshot.backup.rightafter from global configuration
If snapshot.backup.rightafter is set to false (defaults to true), snapshots are
not backed up to secondary storage
