Adjusts ordering and ensures that node health states are included in
faults if they are less than 50%.
Also adjusts fault ID generation and runs fault checks only coordinator
nodes to avoid too many runs.
Moves all tasks run by the Celery worker into a discrete package/module
for easier installation. Also adjusts several parameters throughout to
accomplish this.
Ensures that messages are fully read before each append. Adds more
Zookeeper hits, but ensures logs won't be overwritten by multiple
daemons.
Also don't use a set on the client side, to avoid "removing duplicate"
entries erroneously.
This is still needed due to the nature of the locks and freeing them on
startup, and to preserve lock=fail behaviour on VM startup.
Also fixes the fencing lock flush to directly use the client library
outside of Celery. I don't like this hack but it seems prudent until we
move fencing to the workers as well.
1. Simplify this by leveraging the existing remove_osd/add_osd
functions, since its task was functionally identical to those two in
sequential order.
2. Add support for split OSDs within the command (replacing all OSDs on
the block device(s) as required).
3. Add additional configurability and flexibility around the old device,
weight, and external DB LVs.
Allows creating multiple OSDs on a single (NVMe) block device,
leveraging the "ceph-volume lvm batch" command. Replaces the previous
method of creating OSDs.
Also adds a new ZK item for each OSD indicating if it is split or not.
This helps ensure an easier restore as the tar archive(s) can be sent
directly to the API via the normal process of image uploading, instead
of individual disks.
Converting into human results in imprecise values when specifying bytes
directly, which in turn breaks VMDK image uploads. Instead, just use the
raw bytes value when creating the volume instead of converting it back.
It didn't make any sense to me for mem(prov) to be the default selector,
since this has too many caveats versus mem(free). Switch to using
mem(free) as the default (i.e. "mem") and make memprov the alternative.
Otherwise the node entries could come back in an arbitrary order; since
this is an ordered list of dictionaries that might not be expected by
the API consumers, so ensure it's always sorted.
1. Add documentation on the node selector flags. In the API, reference
the daemon configuration manual which now includes details in this
section; in the CLI, provide the help in "pvc vm define" in detail and
then reference that command's help in the other commands that use this
field.
2. Ensure the naming is consistent in the CLI, using the flag name
"--node-selector" everywhere (was "--selector" for "pvc vm" commands and
"--node-selector" for "pvc provisioner" commands).
Adds commands to both replace an OSD disk, and refresh (reimport) an
existing OSD disk on a new node. This handles the cases where an OSD
disk should be replaced (either due to upgrades or failures) or where a
node is rebuilt in-place and an existing OSD must be re-imported to it.
This should avoid the need to do a full remove/add sequence for either
case.
Also cleans up some aspects of OSD removal that are identical between
methods (e.g. using safe-to-destroy and sleeping after stopping) and
fixes a bug if an OSD does not truly exist when the daemon starts up.
Ensures that information like the FSIDs and the OSD LVM volume are
stored in Zookeeper at creation time and updated at daemon start time
(to ensure the data is populated at least once, or if the /dev/sdX
path changes).
This will allow safer operation of OSD removals and the potential
implementation of re-activation after node replacements.
Allows an administrator to adjust the PG count of a given pool. This can
be used to increase the PGs (for example after adding more OSDs) or
decrease it (to remove OSDs, reduce CPU load, etc.).
Allows specifying a particular device class ("tier") for a given pool,
for instance SSD-only or NVMe-only. This is implemented with Crush
rules on the Ceph side, and via an additional new key in the pool
Zookeeper schema which is defaulted to "default".
Solves two problems:
1. How match fuzziness was used was very inconsistent; make them all the
same, i.e. "if is_fuzzy and limit, apply .* to both sides".
2. Use re.fullmatch instead of re.match to ensure exact matching of the
regex to the value. Without fuzziness, this would sometimes cause
inconsistent behavior, for instance if a limit was non-fuzzy "vm",
expecting to match the actual "vm", but also matching "vm1" too.
Instead of requiring the VM to already be stopped, instead allow disable
state changes to perform a shutdown first. Also add a force option which
will do a hard stop instead of a shutdown.
References #148
Refactors some of the code in VXNetworkInterface to handle MTUs in a
more streamlined fashion. Also fixes a bug whereby bridge client
networks were being explicitly given the cluster dev MTU which might not
be correct. Now adds support for this option explicitly in the configs,
and defaults to 1500 for safety (the standard Ethernet MTU).
Addresses #144
When using the "state", "node", or "tag" arguments to a VM list, add
support for a "negate" flag to look for all VMs *not in* the state,
node, or tag state.
The default of 0.05 (5%) is likely ideal in the initial implementation,
but allow this to be set explicitly for maximum flexibility in
space-constrained or performance-critical use-cases.
Adds in three parts:
1. Create an API endpoint to create OSD DB volume groups on a device.
Passed through to the node via the same command pipeline as
creating/removing OSDs, and creates a volume group with a fixed name
(osd-db).
2. Adds API support for specifying whether or not to use this DB volume
group when creating a new OSD via the "ext_db" flag. Naming and sizing
is fixed for simplicity and based on Ceph recommendations (5% of OSD
size). The Zookeeper schema tracks the block device to use during
removal.
3. Adds CLI support for the new and modified API endpoints, as well as
displaying the block device and DB block device in the OSD list.
While I debated supporting adding a DB device to an existing OSD, in
practice this ended up being a very complex operation involving stopping
the OSD and setting some options, so this is not supported; this can be
specified during OSD creation only.
Closes#142
Ensures that a VM won't:
(a) Have provisioned more RAM than there is available on a given node.
Due to memory overprovisioning, this is simply a "is the VM memory count
more than the node count", and doesn't factor in free or used memory on
a node, total cluster usage, etc. So if a node has 64GB total RAM, the
VM limit is 64GB. It is up to an administrator to ensure sanity *below*
that value.
(b) Have provisioned more vCPUs than there are CPU cores on the node,
minus 2 to account for hypervisor/storage processes. Will ensure there
is no severe CPU contention caused by a single VM having more vCPUs than
there are actual execution threads available.
Closes#139
Adds a new API endpoint to support hot attach/detach of devices, and the
corresponding client-side logic to use this endpoint when doing VM
network/storage add/remove actions.
The live attach is now the default behaviour for these types of
additions and removals, and can be disabled if needed.
Closes#141
This should be managed in ceph.conf with the `rbd default
features` configuration option instead, and thus can be tailored to the
underlying OS version.
If this isn't, the resize will be interpreted as a MB value and result
in an absurdly big volume instead. This is the same consistency
validation that occurs on add.
Convert from human to bytes, then to megabytes and always pass this to
the RBD command. This ensures consistency regardless of what is actually
passed by the user.
We need to do a bit more finagling with the logger on termination to
ensure that all messages are written and the queue drained before
actually terminating.
Adds the ability to send node daemon logs to Zookeeper to facilitate a
command like "pvc node log", similar to "pvc vm log". Each node stores
its logs in a separate tree under "/logs" which can then be combined or
queried. By default, set by config, only 2000 lines are kept.
Add an additional protected class, limit manipulation to one at a time,
and ensure future flexibility. Also makes display consistent with other
VM elements.
Adds tags to schema (v3), to VM definition, adds function to modify
tags, adds function to get tags, and adds tags to VM data output.
Tags will enable more granular classification of VMs based either on
administrator configuration or from automated system events.
Prevents bad states where the VM is "removed" but some of its disks
remain due to e.g. stuck watchers.
Rearrange the sequence so it goes stop, delete disks, then delete VM,
and then return a failure if any of the disk(s) fail to remove, allowing
the task to be rerun after fixing the problem.
Regenerating the ZK connection was fraught with issues, including
duplicate connections, strange failures to reconnect, and various other
wonkiness.
Instead let Kazoo handle states sensibly. Kazoo moves to SUSPENDED state
when it loses connectivity, and stays there indefinitely (based on
cursory tests). And Kazoo seems to always resume from this just fine on
its own. Thus all that hackery did nothing but complicate reconnection.
This therefore turns the listener into a purely informational function,
providing logs of when/why it failed, and we also add some additional
output messages during initial connection and final disconnection.
When trying to write to sub-item paths that don't yet exist, the
previous method would just blindly write to whatever the root key is,
which is never what we actually want.
Instead, check explicitly for a "base path" situation, and handle that.
Then, if we try to get a subpath that isn't valid, return None. Finally
in the various functions, if the path is None, just continue (or return
false/None) and (try to) chug along.
The zkhandler read() function takes care of ensuring there is a None
value returned if these fail, so these aren't required. Makes the code a
fair bit more readable here.
This helps parallelize the numerous Zookeeper calls a little bit, at
least within the bounds of the GIL, to improve performance when getting
a large list of VMs. The max_workers value is capped at 32 to avoid
causing too many threads during concurrent executions, but still
provides a noticeable speedup (on the order of 0.2-0.4 seconds with 75
VMs, scaling up further as counts grow).
Usable anywhere that the global daemon "config" parameter can be passed
in (e.g. pvcapid/helper.py, pvcnoded/Daemon.py, etc.). Stores results in
a subdirectory of the PVC logdir called "profiler" if this directory can
be created, or prints results.
The debug config parameter ensures that the profiler can be added to
functions and not run unless the server is explicitly in debug mode.
Might not be useful as I don't initially plan to add this to every
function (only when investigating performance problems), but this
flexibility allows that to change later.
Instead of looping 5+ times acquiring an impossible lock on a
nonexistent key, just fail on a different error and return failure
immediately.
This is likely a major corner case that shouldn't happen, but better to
be safe than 500.
Joshua Boniface