The new implementation can return it when unable to lock the node.
Other possible errors are 400 and 404 (should not be retried), as well as
5xx (already retried).
Change-Id: I74c2f54a624dc47e8e2d1e67ae4c6a6078e01d2f
IPA can only be run as root and does not use rootwrap. We need to
eventually remove support for rootwrap from ironic-lib.
Change-Id: Iffd5cae5e3dc8637bc6dd10b3bcc9fe33932b8cf
IPA reports a few cpu fields including cores, arch, flags etc.
There is a need that user wants to utilize the physical number in
a baremetal since cores are just a logical representation of the
compute resource.
The socket number is more suitable for the quota control in some
use cases.
Change-Id: I94be86d6b12a3a7e7ca1041d948427a073412a31
With the new in-band inspection, we can derive the callback URL from
the Ironic URL, there is no need to duplicate it. This change uses
the presence of collectors as a sign to run inspection.
The previous approach of setting an inspection URL, with or without
explicitly setting collectors, still works for compatibility with
ironic-inspector.
Change-Id: Ie4279ee6d2995c9686f1dcdef1d6e5dc1dd20871
Allows nodes with a single IP stack to be deployed from a dual-stack
Ironic.
Detecting advertised address and usable Ironic URLs are done completely
independently which does open some space for a misconfiguration. I hope
it's not likely in the reality, especially since this feature is
targetting advanced standalone users.
Change-Id: Ifa506c58caebe00b37167d329b81c166cdb323f2
Closes-Bug: #2045548
Somehow, it has worked correctly for years, but now I've discovered that
the new inspection is (no longer?) tolerant to the missing header.
While here, copy all headers from the heartbeat code.
Change-Id: I9e5c609eb4435e520bc225dea08aedfdf169744b
This fixes several spelling issues identified by codepsell. In some
cases, I may have manually modified a line to make the output more clear
or to correct grammatical issues which were obvious in the codespell
output.
Later changes in this chain will provide the codespell config used to
generate this, as well as adding this commit's SHA, once landed, to a
.git-blame-ignore-revs file to ensure it will not pollute git historys
for modern clients.
Related-Bug: 2047654
Change-Id: I240cf8484865c9b748ceb51f3c7b9fd973cb5ada
Since we moved to exponential wait we increased the amount of time
to run unit tests, now we can configure the max time to wait
- before: Ran: 33 tests in 22.6581 sec.
- after: Ran: 33 tests in 4.0256 sec.
Change-Id: Ibdcfebacad0489d17183e43ceb0d603fce67e72b
Currently, if heartbeat fails, we reschedule it after 5 seconds.
This is fine for the first retry, but it can cause a thundering herd
problem when a lot of nodes fail to heartbeat at once.
This change adds jitter to the minimum wait of 5 seconds. The jitter is
not applied for forced heartbeats: they still have a minimum wait of
exactly 5 seconds from the last heartbeat.
The code is re-ordered to move the interval calculation to one place.
Bonus: correctly logging the next interval.
The unit tests have been rewritten to test the heartbeat process step by
step and not rely on the exact sequence of the calls.
Closes-Bug: #2038438
Change-Id: I4c4207b15fb3d48b55e340b7b3b54af833f92cb5
* ProxyError is derived from ConnectionError, but it's necessary
to check the Response object to identify.
- Added ProxyError in retry_if_exception_type
- Updated _post_to_inspector to proper handle ProxyError
- Updated the wait to use wait_exponential instead of wait_fixed.
Closes-Bug: 2045429
Change-Id: Iefe3fe581cd4e7c91a0da708e6f6d0fdaacab6fe
This reverts commit 33f01fa3c2.
There are a few issues with the patch - see my comments there.
The most pressing and the reasons to revert are:
1) It breaks deployments when the vmedia is present but does not
have a network_data.json (the case for Metal3).
2) It assumes the presence of Glean which may not be the case.
Neither Julia nor myself have time to thoroughly fix the issue,
leaving a revert as the only option to unblock Metal3.
Change-Id: I3f1a18a4910308699ca8f88d8e814c5efa78baee
Closes-Bug: #2045255
This change improves the regex to match an exact entry name, and to also
match with the the entry type from a set of recognised types.
The boot entry details start from the recognised type onwards.
This can be used by a step which deletes all entries of type 'HW' and
UsbClass.
Related-Bug: #2041901
Change-Id: I5d879f724efc2919b541fd3fef0f931df67ff9c7
In some cases the output of the multipath can differ
and we would return a wrong parent device.
Closes-Bug: 2043992
Change-Id: I848d7df798cc736bd5a55eed8fa46110caea1dc3
When performing DHCP-less deployments, the agent can start and
discover more than one configuration drive present on a host.
For example, a host was previously deployed using Ironic, and
is now being re-deployed again.
If Glean was present in the ramdisk, the glean-early.sh would end
mounting the folder based upon label.
If cloud-init, somehow is still in the ramdisk, the other folder
could somehow get mounted.
This patch, which is intended to be backportable, causes the agent
to unmount any configuration drive folders, mount the most likely
candidate based upon device type, partition, and overall state of
the machine, and then utilize that configuration, if present,
to re-configure and reload networking.
Thus allowing dhcp-less re-deployments to be fixed without
forcing any breaking changes.
It should also be noted that this fix was generated in concert
with an additional tempest test case, because this overall failure
case needed to be reproduced to ensure we had a workable non-breaking
path forward.
Closes-Bug: 2032377
Change-Id: I9a3b3dbb9ca98771ce2decf893eba7a4c1890eee
A step will be developed to delete all EFI entries of type HD. As part
of this get_boot_record will need to parse more of the output of
`efibootmgr -v`.
This change asserts the existing behaviour of get_boot_record, and the
test can evolve with the changes in get_boot_record.
Related-Bug: #2041901
Change-Id: I0c5ac4adc1044c528c27a4eaf580c619ceef47e0
Image caching was never fully supported in Ironic or IPA; this is vestigal
code leftover from a partial implementation.
Even if we implemetented it today, we'd likely use a completely different
methodology.
Change-Id: Id4ab7b3c4f106b209585dbd090cdcb229b1daa73
IPA now includes information about numa node id when collecting
information about PCI devices.
Closes-bug: #1622940
Co-Authored-By: Jay Faulkner <jay@jvf.cc>
Change-Id: I70b0cb3eff66d67bb8168982acbbf335de0599cd
This commit:
- Adds the ability to ignore inconsequential OS error caused
by starting the multipathd service when an instance of the
service is already running.
Related launchpad issue https://bugs.launchpad.net/ironic-python-agent/+bug/2031092
Change-Id: Iebf486915bfdc2546451e6b38a450b4c241e43a8
Used helper function, `parse_device_tags`
from ironic_lib instead of the
shlex module for their identical
functionality. Updated
mock_execute.side_effect for lsblk
compatibility in utils.execute.
Closes-Bug: #2037572
Change-Id: I6600e054f9644c67ab003f0e0f6c380b5c217223
HTTP is a fun protocol.
Size is basically optional. And clients implicitly trust the server
and socket has transferred all the bytes. Which *really* means you
should always checksum.
But... previously we didn't checksum as part of retrying.
So if anything happened with python-requests, or lower level
library code or the system itself causing bytes to be lost off the
buffer, creating an incomplete transfer situation, then we wouldn't
know until the checksum.
So now, we checksum and re-trigger the download if there is a
failure of the checksum.
This involved a minor shift in the download logic, and resulted in
a needful minor fix to an image checksum test as it would loop for
90 seconds as well.
Closes-Bug: 2038934
Change-Id: I543a60555a2621b49dd7b6564bd0654a46db2e9a
Follow-up from service steps addition change to add a deploy steps
alias for the Nvidia Mellanox network device firmware update clean
steps. This allows deploy time firmware updates to be codified as
part of a deployment with custom steps.
Change-Id: I9d80447dee7cfde4d3f8d81d9d39e738916b7824
Initial code patches for service steps have merged in
ironic, and it is now time to add support into the
agent which allows service steps to be raised to
the service.
Updates the default hardware manager version to 1.2,
which has *rarely* been incremented due to oversight.
Change-Id: Iabd2c6c551389ec3c24e94b71245b1250345f7a7
When an underlying block device (or driver) only supports 4KB IO,
this can cause some issues with aspects like using an ISO9660 filesystem
which can only support a maximum of 2KB IO.
The agent will now attempt to mount the filesystem *before* deleting the
supplied file, and should that fail it will mount the configuration drive
file from the ramdisk utilizing a loopback, and then extract the contents
of the ramdisk into a newly created VFAT filesystem which supports 4KB
block IO.
Closes-Bug: #2028002
Change-Id: I336acb8e8eb5a02dde2f5e24c258e23797d200ee
If the node is locked, a lookup cannot be performed when an agent
token needs to be generated, which tends to error like this:
ironic_python_agent.ironic_api_client [-] Failed looking up node
with addresses '00:6f:bb:34:b3:4d,00:6f:bb:34:b3:4b' at
https://172.22.0.2:6385. Error 409: Node
c25e451b-d2fb-4168-b690-f15bc8365520 is locked by host 172.22.0.2,
please retry after the current operation is completed..
Check if inspection has completed.
Problem is, if we keep pounding on the door, we can actually worsen
the situation, and previously we would just just let tenacity
retry.
We will now hold for 30 seconds before proceeding, so we have
hopefully allowed the operation to complete.
Also fixes the error logging to help human's sanity.
Change-Id: I97d3e27e2adb731794a7746737d3788c6e7977a0
Rebuilding an instance on a RAIDed ESPs will fail due to sgdisk
running against an non-clean disk and bailing out. Check if there
is a RAIDed ESP already and skip creation if it exists.
Change-Id: I13617ae77515a9d34bc4bb3caf9fae73d5e4e578
When troubleshooting download issues, which may present
as checksum validation failures, it is difficult to understand
if the *entire* file was downloaded due to the way HTTP works.
In that, a download may start with a successful result code,
and the content is streamed out until the socket is closed.
But with HTTP there is no way to know if that socket closed
prematurely and the original server size is *also* an optional
field, so just log the size we got to so we don't drive the
humans [more-]insane.
Also now logs the (optional) content-length field if
supplied by the server.
Change-Id: Id71b167f4e330d54b9afddf95f1a2ef9e40398bf