[sos] Add 'upload' component to upload existing reports and files #3746
Conversation
|
Huge thanks to @pmoravec for all the help reviewing this, suggesting improvements, and finding bugs. |
|
Congratulations! One of the builds has completed. 🍾 You can install the built RPMs by following these steps:
Please note that the RPMs should be used only in a testing environment. |
|
@arif-ali about these ones: sos/upload/init.py:103:0: C0325: Unnecessary parens after '=' keyword (superfluous-parens) ************* Module sos.upload sos/upload/init.py:42:46: W0613: Unused argument 'in_place' (unused-argument) sos/upload/init.py:43:17: W0613: Unused argument 'hook_commons' (unused-argument) sos/upload/init.py:155:24: R1722: Consider using 'sys.exit' instead (consider-using-sys-exit) |
jcastill
force-pushed
the
jcastillo-add-upload-subsystem
branch
from
d5b6c64 to
0e6bc72
Compare
|
With R1725, I made the changes a few months back, and hence enabled the check, so let's do this here too. With the unused variable. If your 100% sure you're going to be using them then potentially you could add the following before the line |
jcastill
force-pushed
the
jcastillo-add-upload-subsystem
branch
from
0e6bc72 to
9c60d66
Compare
Done, should be in the version I just pushed.
Nice one! But I ended up removing it. I'll re-add them in the future when I have ready the code for hooking report etc. |
man/en/sos-upload.1
Outdated
|
|
||
| .PP | ||
| .SH DESCRIPTION | ||
| upload is an sos subcommand to upload sos reports, logs, vmcores, or other files to a policy defined remote location, or an user defined one. |
There was a problem hiding this comment.
Nitpick: s/an user/a user/.
There was a problem hiding this comment.
I think is 'an' because 'user' starts with a vowel, no?
There was a problem hiding this comment.
The rule is "first sound of word", not first letter :) E.g. https://english.stackexchange.com/questions/105116/is-it-a-user-or-an-user (though I am not sure how authoritative that source is).
There was a problem hiding this comment.
Nice, TIL. Fixed in the next push.
sos/policies/distros/redhat.py
Outdated
| @@ -298,7 +298,7 @@ def get_upload_url(self): | |||
| self.ui_log.info("No case id provided, uploading to SFTP") | |||
| return RH_SFTP_HOST | |||
| rh_case_api = "/support/v1/cases/%s/attachments" | |||
| return RH_API_HOST + rh_case_api % self.case_id | |||
| return RH_API_HOST + rh_case_api % self.commons['cmdlineopts'].case_id | |||
There was a problem hiding this comment.
Why this change? AFAIK self.case_id might be blank (and common's case_id set) only in scenario "case id not in cmdline, batch not in cmdline" - should not upload query for case_id, then? (or am I wrong here with my assumption)?
(that concern is valid for sure (while I can be wrong on its impact to this code change):
# python3 bin/sos upload /var/tmp/sosreport-pmoravec-rhel8-012345678-2024-08-13-gbiatgg.tar.xz
sos upload (version 4.7.2)
This utility is used to upload files to a policy-default location.
The archive to be uploaded may contain data considered sensitive and its content
should be reviewed by the originating organization before being passed to any
third party.
No configuration changes will be made to the system running this utility.
Press ENTER to continue, or CTRL-C to quit
Attempting to upload file /var/tmp/sosreport-pmoravec-rhel8-012345678-2024-08-13-gbiatgg.tar.xz to case
No case id provided, uploading to SFTP
No case id provided, uploading to SFTP
Attempting upload to Red Hat Secure FTP
..
There was a problem hiding this comment.
This is one of the things we talked about internally when I first started playing with 'upload'. If you remember, the issue was that without this change, we were getting 'None' on the case_id and it was failing to build the url, and so failed to upload. I have the feeling that I've done something wrong on the upload side and I'm not passing the case_id correctly.
My hope is that more experienced eyes, or at least fresher, can tell me where I'm failing.
|
When I run: and "Press ENTER to continue", and then nothing, then I get a final error: I think the upload did not succeed at the end.. |
| defined location. These files can be either sos reports, | ||
| sos collections, or other kind of files like: vmcores, | ||
| application cores, logs, etc. | ||
|
|
There was a problem hiding this comment.
I don't think this has been addressed, we could just remove the extra line
|
When pressing Ctrl+C on |
|
@pmoravec thank you for finding this, I thought we solved these issues:
I'll check the double messaging here, looks horrible.
I'll check this one as well, I remember we had a similar issue with a previous implementation.
No, it should not succeed in that case. |
I'll check this, should be easy to fix. |
Fixed. I used exit() instead of _exit(), which is the one implemented in Soscomponent. |
There was a problem hiding this comment.
At a bare minimum, a new component should be implementing all the abstractions that it needs to operate solo, not acting as a wrapper to existing functionality.
This means the upload logic needs to be separated from its current location in Policy, and implemented as a discrete unit. Policy should then control the default setting, and users should be able to direct sos to chose an upload target/profile/whatever we want to call it as an override. E.G. if I have an sos report locally on my Fedora workstation that was taken from a RHEL box, and I am unable due to some network policy to directly upload from the RHEL box, then on my Fedora system I should be able to send that sos report to Red Hat.
Further, any current or future usage of the component's functionality should go through the actual component code. Much like we do with sos clean, when --clean is used for a report being generated. We hook into the component from within report, to ensure we use the exact code flow for cleaning the archive as we would by running a clean after-the-fact.
jcastill
force-pushed
the
jcastillo-add-upload-subsystem
branch
from
9c60d66 to
72dd27c
Compare
I agree with everything above, but the idea behind this PR is to be a first implementation to get the upload component started, and then move things carefully from policy to upload. Could this approach be acceptable? |
|
I support this initial implementation of the feature to let enhance I was thinking to raise the same concern, but I realized I would see beneficial for the discussion about refactorization if we already has some implementation in hand. With the current code, it is hard for me to specify "cut this away from here and put it (there)", if we have no "(there)". With the If somebody sees as a potential threat "we accept this initial implementation, but will never refactor the code as needed, and we dont want that technical debt here", then I can make a commitment: once there will be an agreement about the refactorisation and if nobody(*) will have time to implement it, I will work on such PR. (*) nobody including Jose as the primary person to implement. I assume he will be the primary person to make his own feature to make it complete. On the other side, there can be various reasons he won't be able to do the refactorisation (time, other work on sos, willingness, whatever). And then anybody else (with me as the volunteer with above commitment) can contribute that way. |
On this note, I already started moving things around from policies/distros just after I sent this PR - this is not something I want to leave abandoned, or done in six months time or more, but as soon as possible. But also I want to make sure I cover all the possible cases, and the upload code in policies has been there for a long time, working perfectly, so want to be extra careful while refactoring. |
| [--case-id id]\fR | ||
| [--upload-url url]\fR | ||
| [--upload-user user]\fR | ||
| [--upload-pass pass]\fR | ||
| [--upload-directory dir]\fR | ||
| [--upload-method]\fR | ||
| [--upload-no-ssl-verify]\fR | ||
| [--upload-protocol protocol]\fR |
There was a problem hiding this comment.
@jcastill Could the --upload-protocol s3 flags be included in this work? Unfortunately, it contains unique flags that made S3 easier to implement at the time.
[--upload-s3-endpoint endpoint]
[--upload-s3-region region]
[--upload-s3-bucket bucket]
[--upload-s3-access-key access_key]
[--upload-s3-secret-key secret_key]
[--upload-s3-object-prefix object_prefix]
The existing flags and how the provided values were used were not well aligned for S3, even though valid for FTP, HTTP, etc. protocols. I didn't want to cause any breakage for existing upload protocols while trying to make them work for all protocols, so S3 ended up with unique flags.
I planned to attempt a refactor at some point (sos v5?) where the original protocols and s3 overlap. For example, allowing synonymous flags:
--upload-user~--upload-s3-access-key--upload-pass~--upload-s3-secret-key--upload-directory~--upload-s3-object-prefix--upload-url~--upload-s3-endpoint
However, I haven't been able to dedicate the time yet.
There was a problem hiding this comment.
Could the --upload-protocol s3 flags be included in this work? Unfortunately, it contains unique flags that made S3 easier to implement at the time.
Yes, I'll make sure I include them in the next iteration of this PR.
I planned to attempt a refactor at some point (sos v5?) where the original protocols and s3 overlap. For example, allowing synonymous flags:
--upload-user ~ --upload-s3-access-key --upload-pass ~ --upload-s3-secret-key --upload-directory ~ --upload-s3-object-prefix --upload-url ~ --upload-s3-endpointHowever, I haven't been able to dedicate the time yet.
Let me know if I can help. My original idea was to have this PR as a starting point and then move stuff out of the generic policy and the OS-specific ones in a second PR, but that was rejected, so I'm working on the full change now. As soon as I finish with that, we can start working on the refactor of S3 it that's OK with you. In fact we need to do some work with S3 uploads for the RH customer portal, so we could do both things in parallel.
There was a problem hiding this comment.
Let me know if I can help. My original idea was to have this PR as a starting point and then move stuff out of the generic policy and the OS-specific ones in a second PR, but that was rejected, so I'm working on the full change now. As soon as I finish with that, we can start working on the refactor of S3 it that's OK with you. In fact we need to do some work with S3 uploads for the RH customer portal, so we could do both things in parallel.
When you have a branch published for public view and somewhat functional let me know. I'll branch off of it and start migrating the s3 portions in then submit a PR targeting your branch for you to review.
As for the s3 refactoring, we can look into it and I'd be more than happy to try and make some time. I believe a few lend themselves easily, or at least I don't immediately recall any issues with using them, like URL, user, and password. One I do recall bringing up some questions is the --upload-directory. For example, should this be only the prefixes inside the bucket? Or should it split the directory like {bucket}/{prefix} on only the first slash? There may have been others, but I would have to review the LinuxPolicy.get_upload_xxxx() functions and internal self._vars again.
Without some "group think" I decided not to implement something I (or others) may have been unhappy with later but stuck with unless making breaking changes.
There was a problem hiding this comment.
There may be less to refactor than I first thought as I haven't reviewed the code in almost a year. I guess ended up implementing some of it already. Hope I'm still happy with my choices after a year 😄
sos/sos/policies/distros/__init__.py
Lines 615 to 629 in 2aa4fcf
arif-ali
added
the
Status/Need More Info
|
Latest push:
|
Apologies for delay, I was travelling to the EU when the comments on s3 were made. I'll try to test this PR for s3 uploads later today, or before Monday at the latest. |
|
Thank you so much @TrevorBenson . I think we can wait until Monday, so you can enjoy your trip to Europe :) |
|
Tests from Fedora: And from RHEL: Both fail as expected because I didn't have the right credentials, but I wanted to check that the actual code flow was working. |
|
I had some issues after installing dnf-plugins-core, trying to enable packit/sosreport-sos-3746 fails I meant to get back to looking into this sooner, but had not had much free time. I'll give it another shot this evening to see if I can isolate why the Fedora 41 system is failing to install the packit RPM. |
|
It seems we set up just rawhide? Since I fail the same also on F40 , while I see https://download.copr.fedorainfracloud.org/results/packit/sosreport-sos-3746/ exists. And indeed, commands from https://dashboard.packit.dev/jobs/copr/2168507 do work: |
|
Is the |
I think this may be a consecuence of my change of target names. Let me fix it and push a new commit |
jcastill
force-pushed
the
jcastillo-add-upload-subsystem
branch
from
6e884c6 to
06245ce
Compare
|
@TrevorBenson Try with target GenericUpload, i.e.:
Changing whatever you need from there. I added GenericUpload explicitly, so 'local' will be used when the person running the command wants to detect the local target (and that is the default). |
|
Wait...clarify something for me, as I don't have a RHEL box handy to explicitly test this with at the moment. Is the current design in this PR:
Side note - I'd like it if, regardless of the answer to the above, users don't specify |
First option. So users can do: a- Upload from RHEL, either: So when I tested uploads from RHEL I did an "implicit target" upload, i.e. not specifying; or specifying the others (so, weirdcase but still, upload to Ubuntu target from RHEL).
I dislike *Upload (RHELUpload, UbuntuUpload, GenericUpload) as target names inmensely, so happy to change that as soon as possible to something that makes sense and doesn't get users' fingers exhausted. I imagine "redhat", "ubuntu", "generic", "local" could be better. |
|
ACK to Jose's comment. Targets work like presets, automatically loaded but can be overriden, I verified that as well. ACK to the rename of targets to e.g. |
|
Actually, self selection of profile works well for existing profiles but not for non-explicit ones, so when @TrevorBenson ran his tests in Fedora, it wasn't selecting any,, and it has to do with one of the changes I did when moved from static to dynamic class selection for upload. |
👍🏼 I've got a 6 hour layover in Munich tomorrow where I can focus on this testing. |
What I'm suggesting is that the class names are fine, but the strings the end users leverage should be friendlier. Like we do with plugins with the Like Pavel I prefer |
Yes, I meant the same, only user-facing options, I should have been clearer. This is how it looks now:
This next version I'm trying to push at the moment still has 'ubuntu', but I can change it to 'canonical' pretty quickly. |
jcastill
force-pushed
the
jcastillo-add-upload-subsystem
branch
from
06245ce to
ace6660
Compare
@TrevorBenson thank you so much for testing this, I really appreciate you taking the time to do it. |
|
I tested RPM @jcastill LGTM from my side on S3 protocol uploads. All tests were using amazon AWS S3. I did find on bug while testing. However, I went ahead and went through comparisons to version 8 and 9 enterprise distributions to make sure they were still working as intended. I've opened #3921 for tracking purposes to try and work out a fix before the packages make it into enterprise and the error is observed. FWIW, I didn't find a CentOS Stream 9 container to perform testing with and downloading the Stream 9 ISO was a bit slow in the airport. So I skipped testing for this PR except for F41. Given a packit repository for centos-stream-8, to go along with stream 9, I would test under both stream versions as well as F41 for any future s3, saltstack or other major changes I provide testing for. As sos has approval to be upgraded after Enhancements are no longer being applied to other packages in RHEL 8 it would be my preference to test thoroughly since they have diverged packages versions and when sos is updated, python3-boto3, botocore, etc. may not be getting updated as well, so creates a small possibility of testing only on rhel9/f41 in missing something that could impact rhel8. |
I think |
sos/upload/targets/__init__.py
Outdated
| """ | ||
|
|
||
| desc = """ | ||
| Upload a file (can be a sos report, a must-gather, or others) to |
There was a problem hiding this comment.
Super minor nit that I'm only highlighting since this string can be user facing: this should be "an sos report", as it is officially "ess oh ess" and not "sauce".
This commit marks the beginning of the addition of a new 'upload' component for sos, which can be used to upload already created sos reports, collects, or other files like logs or vmcores to a policy defined location. The user needs to specify a file location, and can make use of any of the options that exist nowadays for the --upload option. This first commit includes: - The initial framework for the 'upload' component. - The new man page for 'sos upload'. - The code in the component 'help' to show information about the component. - The code in sos/__init__.py to deal with the component. - The code for uploads to Red Hat and Ubuntu systems. - The code to allow uploads specifying remote destination, called targets in this implementation. For example, you could generate a sos report in a CentOS system and specify a target defined as 'redhat' or 'RedHatUpload' to upload to the Red Hat Customer Portal. - And modifications to setup.py to build the man pages. Related: RHEL-23032, SUPDEV-138, CLIOT-481 Co-authored-by: Jose Castillo <jcastillo@redhat.com> Co-authored-by: Pavel Moravec <pmoravec@redhat.com> Co-authored-by: Trevor Benson <trevor.benson@gmail.com> Signed-off-by: Jose Castillo <jcastillo@redhat.com>
jcastill
force-pushed
the
jcastillo-add-upload-subsystem
branch
from
ace6660 to
ab15651
Compare
|
Thank you so much everyone for reviewing this PR and for all your suggestions. I moved the option name from 'ubuntu' to 'canonical' and fixed the 'a sos to 'an sos'. If there's nothing else, can you merged this PR? |
This commit marks the beginning of the addition of a new 'upload' component for sos, which can be used to upload already created sos reports, collects, or other files like logs or vmcores to a policy defined location.
The user needs to specify a file location, and can make use of any of the options that exist nowadays for the --upload option.
This first commit includes:
Related: RHEL-23032, SUPDEV-138, CLIOT-481
Please place an 'X' inside each '[]' to confirm you adhere to our Contributor Guidelines