* Email based retest request process: proposal for new pull/re-apply feature @ 2024-02-20 15:21 Patrick Robb 2024-02-20 18:12 ` Aaron Conole 0 siblings, 1 reply; 11+ messages in thread From: Patrick Robb @ 2024-02-20 15:21 UTC (permalink / raw) To: ci; +Cc: dev, Aaron Conole, zhoumin [-- Attachment #1: Type: text/plain, Size: 1990 bytes --] Hi all, I want to poll the CI group and dev community about a proposed feature addition to the CI retest request framework. Currently, you can respond to a patchseries or patch email, requesting a retest like so: Recheck-request: iol-compile-amd64-testing, iol-unit-amd64-testing Labs who have added this functionality (UNH and the GitHub Robot) will then trigger retests according to the contexts provided, using the ORIGINAL dpdk artifact they produced at the time when the patch was submitted. This is useful for requesting a retest on a patch when you believe a failure may have been an infra failure or spurious. It is not useful if you believe the tree your patch was applied on was in a bad state when your patch was submitted, and you would like for your patch to be re-applied on the current tip of the branch. A few people have suggested we add this feature (re-apply to tip of branch and retest). So, we should probably add an option allowing people to indicate they want this behavior instead of the "default" retest. Ferruh emailed me about this a while ago and proposed the following syntax, which I am okay with: Recheck-request,attribute=value: ... So a practical example would look like: Recheck-request,pull=True: iol-sample-apps-testing, iol-compile-amd64-testing, github-robot Also, I believe that although we should still require people to include the contexts they're requesting a retest for for posterity (so we can refer back to it), I think if someone includes the pull keyword, ALL labs should trigger retests for ALL tests. The reason for this is I don't think we should display results side by side on a patchseries which are coming from distinct DPDK artifacts. Readers may assume (rightly, in my opinion) that when they're looking at a results table for a patchseries, those results are all coming from identical DPDK artifacts, and not from distinct DPDK artifacts produced at different times, from different commits. What do you all think? Thanks. [-- Attachment #2: Type: text/html, Size: 2178 bytes --] ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Email based retest request process: proposal for new pull/re-apply feature 2024-02-20 15:21 Email based retest request process: proposal for new pull/re-apply feature Patrick Robb @ 2024-02-20 18:12 ` Aaron Conole 2024-02-20 18:24 ` Patrick Robb 0 siblings, 1 reply; 11+ messages in thread From: Aaron Conole @ 2024-02-20 18:12 UTC (permalink / raw) To: Patrick Robb; +Cc: ci, dev, zhoumin Patrick Robb <probb@iol.unh.edu> writes: > Hi all, > > I want to poll the CI group and dev community about a proposed feature addition to the CI retest request framework. > Currently, you can respond to a patchseries or patch email, requesting a retest like so: > > Recheck-request: iol-compile-amd64-testing, iol-unit-amd64-testing > > Labs who have added this functionality (UNH and the GitHub Robot) will then trigger retests according to the contexts > provided, using the ORIGINAL dpdk artifact they produced at the time when the patch was submitted. > > This is useful for requesting a retest on a patch when you believe a failure may have been an infra failure or spurious. It > is not useful if you believe the tree your patch was applied on was in a bad state when your patch was submitted, and > you would like for your patch to be re-applied on the current tip of the branch. A few people have suggested we add > this feature (re-apply to tip of branch and retest). So, we should probably add an option allowing people to indicate they > want this behavior instead of the "default" retest. > > Ferruh emailed me about this a while ago and proposed the following syntax, which I am okay with: > > Recheck-request,attribute=value: ... > > So a practical example would look like: > > Recheck-request,pull=True: iol-sample-apps-testing, iol-compile-amd64-testing, github-robot > > Also, I believe that although we should still require people to include the contexts they're requesting a retest for for > posterity (so we can refer back to it), I think if someone includes the pull keyword, ALL labs should trigger retests for > ALL tests. The reason for this is I don't think we should display results side by side on a patchseries which are coming > from distinct DPDK artifacts. Readers may assume (rightly, in my opinion) that when they're looking at a results table > for a patchseries, those results are all coming from identical DPDK artifacts, and not from distinct DPDK artifacts > produced at different times, from different commits. > > What do you all think? Thanks. Why not something like: Recheck-request: [attribute-list],[test-list]... For example, then we can do: Recheck-request: rebase=[identifier],.... where identifier is a branch specifier (or the word 'latest')? That lets us fixup if the branch picker script guessed a wrong branch. Just spit-balling on syntax. That said, I agree - if a rebase has been requested, all tests need to be rerun. Maybe we should consider that the test labels should be added with a run number or something? Or we could also include that the run is a rerun. That way for labs that don't currently support the recheck request framework, we can easily tell that they weren't re-tried. ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Email based retest request process: proposal for new pull/re-apply feature 2024-02-20 18:12 ` Aaron Conole @ 2024-02-20 18:24 ` Patrick Robb 2024-03-01 14:36 ` zhoumin 0 siblings, 1 reply; 11+ messages in thread From: Patrick Robb @ 2024-02-20 18:24 UTC (permalink / raw) To: Aaron Conole; +Cc: ci, dev, zhoumin [-- Attachment #1: Type: text/plain, Size: 1805 bytes --] On Tue, Feb 20, 2024 at 1:12 PM Aaron Conole <aconole@redhat.com> wrote: > > Why not something like: > > Recheck-request: [attribute-list],[test-list]... > > For example, then we can do: > > Recheck-request: rebase=[identifier],.... > > where identifier is a branch specifier (or the word 'latest')? > I hadn't thought about the option of allowing branch specifiers. Agree that allowing a human correction process for the pw_maintainer_cli.py script choosing the wrong branch sounds helpful. My original idea was offering 2 options (test original artifact, or re-apply on latest). Do we want to support for checking out to a specific commit and re-applying there? I figured that would not be worth it (too much of a niche case), but your comments are making me reconsider. > > That lets us fixup if the branch picker script guessed a wrong branch. > > Just spit-balling on syntax. > > > That said, I agree - if a rebase has been requested, all tests need to > be rerun. Maybe we should consider that the test labels should be added > with a run number or something? Or we could also include that the run > is a rerun. That way for labs that don't currently support the recheck > request framework, we can easily tell that they weren't re-tried. > > so re-report with a modified test label? That is good in that it shows the behavior more clearly. But, it also means we will not overwrite any fails. So the fail will still be there, and the patchwork patch page will grow a huge table. Maybe this is fine. Also raises the point of getting more coverage for the retest framework at other labs. I will email Min Zhou regarding how he uses the dpdk-ci project for the loongson build jobs and see how well that can integrate with the get_reruns.py script. [-- Attachment #2: Type: text/html, Size: 2405 bytes --] ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Email based retest request process: proposal for new pull/re-apply feature 2024-02-20 18:24 ` Patrick Robb @ 2024-03-01 14:36 ` zhoumin 2024-03-04 15:21 ` Aaron Conole 0 siblings, 1 reply; 11+ messages in thread From: zhoumin @ 2024-03-01 14:36 UTC (permalink / raw) To: Patrick Robb, Aaron Conole; +Cc: ci, dev [-- Attachment #1: Type: text/plain, Size: 2800 bytes --] On Wed, Feb 21, 2024 at 2:24AM, Patrick Robb wrote: > > On Tue, Feb 20, 2024 at 1:12 PM Aaron Conole <aconole@redhat.com > <mailto:aconole@redhat.com>> wrote: > > > Why not something like: > > Recheck-request: [attribute-list],[test-list]... > > For example, then we can do: > > Recheck-request: rebase=[identifier],.... > > where identifier is a branch specifier (or the word 'latest')? > > I hadn't thought about the option of allowing branch specifiers. Agree > that allowing a human correction process for the pw_maintainer_cli.py > script choosing the wrong branch sounds helpful. > > My original idea was offering 2 options (test original artifact, or > re-apply on latest). Do we want to support for checking out to a > specific commit and re-applying there? I figured that would not be > worth it (too much of a niche case), but your comments are making me > reconsider. I agree with you that allowing developers to correct the target branch is useful. But, the developer should just provide the name of branch instead of commit ID, which is more reasonable. Of course, the rebasing option is more important. So, I consider we can allow developers to submit a request as following format: Recheck-request: rebase=True|branch=main|contexts=iol-compile-amd64-testing, iol-broadcom-Performance,... We can use "|" as the separator, for example. `rebase` and `branch` can be optional and we can use the default values if the developer doesn't provide them. The default is not rebasing for `rebase` option. The default is the branch chosen by pw_maintainer_cli.py script for `branch` option. The `contexts` option is required. > > Just spit-balling on syntax. > > > That said, I agree - if a rebase has been requested, all tests need to > be rerun. Maybe we should consider that the test labels should be > added > with a run number or something? Or we could also include that the run > is a rerun. That way for labs that don't currently support the > recheck > request framework, we can easily tell that they weren't re-tried. > > so re-report with a modified test label? That is good in that it shows > the behavior more clearly. But, it also means we will not overwrite > any fails. So the fail will still be there, and the patchwork patch > page will grow a huge table. Maybe this is fine. > Re-report with a modified test label may be better. That can tell people more information about the CI testings, such as that the retest indeed happened. > Also raises the point of getting more coverage for the retest > framework at other labs. I will email Min Zhou regarding how he uses > the dpdk-ci project for the loongson build jobs and see how well that > can integrate with the get_reruns.py script. > [-- Attachment #2: Type: text/html, Size: 4898 bytes --] ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Email based retest request process: proposal for new pull/re-apply feature 2024-03-01 14:36 ` zhoumin @ 2024-03-04 15:21 ` Aaron Conole 2024-03-07 17:06 ` Adam Hassick 0 siblings, 1 reply; 11+ messages in thread From: Aaron Conole @ 2024-03-04 15:21 UTC (permalink / raw) To: zhoumin; +Cc: Patrick Robb, ci, dev zhoumin <zhoumin@loongson.cn> writes: > On Wed, Feb 21, 2024 at 2:24AM, Patrick Robb wrote: > > On Tue, Feb 20, 2024 at 1:12 PM Aaron Conole <aconole@redhat.com> wrote: > > Why not something like: > > Recheck-request: [attribute-list],[test-list]... > > For example, then we can do: > > Recheck-request: rebase=[identifier],.... > > where identifier is a branch specifier (or the word 'latest')? > > I hadn't thought about the option of allowing branch specifiers. Agree that allowing a human correction process for > the pw_maintainer_cli.py script choosing the wrong branch sounds helpful. > > My original idea was offering 2 options (test original artifact, or re-apply on latest). Do we want to support for > checking out to a specific commit and re-applying there? I figured that would not be worth it (too much of a niche > case), but your comments are making me reconsider. > > I agree with you that allowing developers to correct the target branch is useful. But, the developer should just provide > the name of branch instead of commit ID, which is more reasonable. Of course, the rebasing option is more important. > So, I consider we can allow developers to submit a request as following format: > > Recheck-request: rebase=True|branch=main|contexts=iol-compile-amd64-testing, iol-broadcom-Performance,... > > We can use "|" as the separator, for example. `rebase` and `branch` can be optional and we can use the default values > if the developer doesn't provide them. The default is not rebasing for `rebase` option. The default is the branch chosen > by pw_maintainer_cli.py script for `branch` option. The `contexts` option is required. Interesting approach. But I don't know about contexts= or something like that. It means there are two passes through the regex. Also, I don't know about contexts either - if the series was requested to rebase, every lab that can re-test probably should since the results aren't going to be valid from the old tests. > Just spit-balling on syntax. > > That said, I agree - if a rebase has been requested, all tests need to > be rerun. Maybe we should consider that the test labels should be added > with a run number or something? Or we could also include that the run > is a rerun. That way for labs that don't currently support the recheck > request framework, we can easily tell that they weren't re-tried. > > so re-report with a modified test label? That is good in that it shows the behavior more clearly. But, it also means > we will not overwrite any fails. So the fail will still be there, and the patchwork patch page will grow a huge table. > Maybe this is fine. > > Re-report with a modified test label may be better. That can tell people more information about the CI testings, such as > that the retest indeed happened. Just back from PTO - actually, I don't think we need to adjust the label, but rather the description. That would allow the mechanism that overwrites the existing test to keep the "checks" page tidy, but also making the retest information clear. WDYT? > Also raises the point of getting more coverage for the retest framework at other labs. I will email Min Zhou > regarding how he uses the dpdk-ci project for the loongson build jobs and see how well that can integrate with the > get_reruns.py script. That would be great! ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Email based retest request process: proposal for new pull/re-apply feature 2024-03-04 15:21 ` Aaron Conole @ 2024-03-07 17:06 ` Adam Hassick 2024-03-18 15:59 ` Patrick Robb 0 siblings, 1 reply; 11+ messages in thread From: Adam Hassick @ 2024-03-07 17:06 UTC (permalink / raw) To: Aaron Conole; +Cc: zhoumin, Patrick Robb, ci, dev On Mon, Mar 4, 2024 at 10:22 AM Aaron Conole <aconole@redhat.com> wrote: > > zhoumin <zhoumin@loongson.cn> writes: > > > On Wed, Feb 21, 2024 at 2:24AM, Patrick Robb wrote: > > > > On Tue, Feb 20, 2024 at 1:12 PM Aaron Conole <aconole@redhat.com> wrote: > > > > Why not something like: > > > > Recheck-request: [attribute-list],[test-list]... > > > > For example, then we can do: > > > > Recheck-request: rebase=[identifier],.... > > > > where identifier is a branch specifier (or the word 'latest')? > > > > I hadn't thought about the option of allowing branch specifiers. Agree that allowing a human correction process for > > the pw_maintainer_cli.py script choosing the wrong branch sounds helpful. > > > > My original idea was offering 2 options (test original artifact, or re-apply on latest). Do we want to support for > > checking out to a specific commit and re-applying there? I figured that would not be worth it (too much of a niche > > case), but your comments are making me reconsider. > > > > I agree with you that allowing developers to correct the target branch is useful. But, the developer should just provide > > the name of branch instead of commit ID, which is more reasonable. Of course, the rebasing option is more important. > > So, I consider we can allow developers to submit a request as following format: > > > > Recheck-request: rebase=True|branch=main|contexts=iol-compile-amd64-testing, iol-broadcom-Performance,... > > > > We can use "|" as the separator, for example. `rebase` and `branch` can be optional and we can use the default values > > if the developer doesn't provide them. The default is not rebasing for `rebase` option. The default is the branch chosen > > by pw_maintainer_cli.py script for `branch` option. The `contexts` option is required. > > Interesting approach. But I don't know about contexts= or something > like that. It means there are two passes through the regex. > > Also, I don't know about contexts either - if the series was requested > to rebase, every lab that can re-test probably should since the results > aren't going to be valid from the old tests. I'm not opposed to having the contexts be a key-value pair argument like the others, however that does break backwards compatibility with our existing syntax. If we don't care very much about backwards compatibility, then we could make this change. Instead of having a boolean and a string parameter for whether to rebase and the branch to rebase on, we could have a single argument specifying a branch. Then, labs rebase on the given branch and then rerun all tests if the "rebase=<branch>" argument is present. This would look like: Recheck-request: rebase=main, iol-sample-apps-testing, iol-unit-amd64-testing, iol-broadcom-Performance Or, using zhoumin's proposed format: Recheck-request: rebase=main|contexts=iol-sample-apps-testing, iol-unit-amd64-testing, iol-broadcom-Performance I don't think the context should be required if the request includes the rebase argument, because we do not want to mix valid and invalid test results as Aaron said. This would be a valid format if contexts are optional: Recheck-request: rebase=main > > Just spit-balling on syntax. > > > > That said, I agree - if a rebase has been requested, all tests need to > > be rerun. Maybe we should consider that the test labels should be added > > with a run number or something? Or we could also include that the run > > is a rerun. That way for labs that don't currently support the recheck > > request framework, we can easily tell that they weren't re-tried. > > > > so re-report with a modified test label? That is good in that it shows the behavior more clearly. But, it also means > > we will not overwrite any fails. So the fail will still be there, and the patchwork patch page will grow a huge table. > > Maybe this is fine. > > > > Re-report with a modified test label may be better. That can tell people more information about the CI testings, such as > > that the retest indeed happened. > > Just back from PTO - actually, I don't think we need to adjust the > label, but rather the description. That would allow the mechanism that > overwrites the existing test to keep the "checks" page tidy, but also > making the retest information clear. WDYT? Yes, if the Community Lab posted new contexts for every retest then the checks page would get quite long. > > Also raises the point of getting more coverage for the retest framework at other labs. I will email Min Zhou > > regarding how he uses the dpdk-ci project for the loongson build jobs and see how well that can integrate with the > > get_reruns.py script. > > That would be great! > ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Email based retest request process: proposal for new pull/re-apply feature 2024-03-07 17:06 ` Adam Hassick @ 2024-03-18 15:59 ` Patrick Robb 2024-03-19 8:36 ` zhoumin 0 siblings, 1 reply; 11+ messages in thread From: Patrick Robb @ 2024-03-18 15:59 UTC (permalink / raw) To: Adam Hassick; +Cc: Aaron Conole, zhoumin, ci, dev On Thu, Mar 7, 2024 at 12:06 PM Adam Hassick <ahassick@iol.unh.edu> wrote: > > > I'm not opposed to having the contexts be a key-value pair argument > like the others, however that does break backwards compatibility with > our existing syntax. If we don't care very much about backwards > compatibility, then we could make this change. > > Instead of having a boolean and a string parameter for whether to > rebase and the branch to rebase on, we could have a single argument > specifying a branch. Then, labs rebase on the given branch and then > rerun all tests if the "rebase=<branch>" argument is present. This > would look like: > > Recheck-request: rebase=main, iol-sample-apps-testing, > iol-unit-amd64-testing, iol-broadcom-Performance I agree with this approach because it preserves backward compatibility, while still providing us with all the functionality we need. We will also be able to accept key value arguments in the future if further feature requests come in which require it. > I don't think the context should be required if the request includes > the rebase argument, because we do not want to mix valid and invalid > test results as Aaron said. > This would be a valid format if contexts are optional: > > Recheck-request: rebase=main Okay, I agree that contexts should not be considered by labs when we use rebase - but of course we will still store the contexts (if they are submitted) alongside the key value args. In the future there may be an application for this. Zhoumin, does this sound acceptable, or do you think there are any flaws? If it works, we will implement the updates and try to upstream this week. Thanks! ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Email based retest request process: proposal for new pull/re-apply feature 2024-03-18 15:59 ` Patrick Robb @ 2024-03-19 8:36 ` zhoumin 2024-03-19 17:30 ` Patrick Robb 0 siblings, 1 reply; 11+ messages in thread From: zhoumin @ 2024-03-19 8:36 UTC (permalink / raw) To: Patrick Robb, Adam Hassick; +Cc: Aaron Conole, ci, dev On Mon, Mar 18, 2024 at 3:59PM, Patrick Robb wrote: > On Thu, Mar 7, 2024 at 12:06 PM Adam Hassick <ahassick@iol.unh.edu> wrote: >> >> I'm not opposed to having the contexts be a key-value pair argument >> like the others, however that does break backwards compatibility with >> our existing syntax. If we don't care very much about backwards >> compatibility, then we could make this change. >> >> Instead of having a boolean and a string parameter for whether to >> rebase and the branch to rebase on, we could have a single argument >> specifying a branch. Then, labs rebase on the given branch and then >> rerun all tests if the "rebase=<branch>" argument is present. This >> would look like: >> >> Recheck-request: rebase=main, iol-sample-apps-testing, >> iol-unit-amd64-testing, iol-broadcom-Performance > I agree with this approach because it preserves backward > compatibility, while still providing us with all the functionality we > need. We will also be able to accept key value arguments in the future > if further feature requests come in which require it. > >> I don't think the context should be required if the request includes >> the rebase argument, because we do not want to mix valid and invalid >> test results as Aaron said. >> This would be a valid format if contexts are optional: >> >> Recheck-request: rebase=main > Okay, I agree that contexts should not be considered by labs when we > use rebase - but of course we will still store the contexts (if they > are submitted) alongside the key value args. In the future there may > be an application for this. > > Zhoumin, does this sound acceptable, or do you think there are any > flaws? If it works, we will implement the updates and try to upstream > this week. Thanks! Thanks for your hard work. I also agree with this approach. The meaning of the key value `rebase=main` is sufficient, and loongson lab can support it. One more thing I want to confirm is whether we should apply the patch onto the branch commit which existed at the time when that patch was submitted or onto the latest tip of branch if users request doing rebase. Users probably request a recheck with `rebase` when the CI lab chose a wrong branch onto which apply the patch. I worry we may encounter conflicts when apply the patch onto the latest commit of the target branch if that branch is just updated before the request. ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Email based retest request process: proposal for new pull/re-apply feature 2024-03-19 8:36 ` zhoumin @ 2024-03-19 17:30 ` Patrick Robb 2024-03-19 17:53 ` Aaron Conole 2024-03-20 1:35 ` zhoumin 0 siblings, 2 replies; 11+ messages in thread From: Patrick Robb @ 2024-03-19 17:30 UTC (permalink / raw) To: zhoumin; +Cc: Adam Hassick, Aaron Conole, ci, dev On Tue, Mar 19, 2024 at 4:37 AM zhoumin <zhoumin@loongson.cn> wrote: > > > One more thing I want to confirm is whether we should apply the patch > onto the branch commit which existed at the time when that patch was > submitted or onto the latest tip of branch if users request doing > rebase. Users probably request a recheck with `rebase` when the CI lab > chose a wrong branch onto which apply the patch. I worry we may > encounter conflicts when apply the patch onto the latest commit of the > target branch if that branch is just updated before the request. > > That's a good edge case to think about... but I also think if the patch no longer applies cleanly on tip of intended branch, then we would be correct to report an apply failure there. And then the submitter should refactor their patch so it applies, and submit again. So I think the process is like A) If retest is requested without rebase key, then retest "original" dpdk artifact (either by re-using the existing tarball (unh lab) or tracking the commit from submit time and re-applying onto dpdk at that commit (loongson)). B) If rebase key is included, apply to tip of the indicated branch. If, because the branch has changed, the patch no longer applies, then we can report an apply failure. Then, submitter has to refactor their patch and resubmit. In either case, report the new results with an updated test result in the email (i.e. report "_Testing PASS RETEST #1" instead of "_Testing PASS" in the email body). ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Email based retest request process: proposal for new pull/re-apply feature 2024-03-19 17:30 ` Patrick Robb @ 2024-03-19 17:53 ` Aaron Conole 2024-03-20 1:35 ` zhoumin 1 sibling, 0 replies; 11+ messages in thread From: Aaron Conole @ 2024-03-19 17:53 UTC (permalink / raw) To: Patrick Robb; +Cc: zhoumin, Adam Hassick, ci, dev Patrick Robb <probb@iol.unh.edu> writes: > On Tue, Mar 19, 2024 at 4:37 AM zhoumin <zhoumin@loongson.cn> wrote: >> >> >> One more thing I want to confirm is whether we should apply the patch >> onto the branch commit which existed at the time when that patch was >> submitted or onto the latest tip of branch if users request doing >> rebase. Users probably request a recheck with `rebase` when the CI lab >> chose a wrong branch onto which apply the patch. I worry we may >> encounter conflicts when apply the patch onto the latest commit of the >> target branch if that branch is just updated before the request. >> >> > > That's a good edge case to think about... but I also think if the > patch no longer applies cleanly on tip of intended branch, then we > would be correct to report an apply failure there. And then the > submitter should refactor their patch so it applies, and submit again. +1 > So I think the process is like > > A) If retest is requested without rebase key, then retest "original" > dpdk artifact (either by re-using the existing tarball (unh lab) or > tracking the commit from submit time and re-applying onto dpdk at that > commit (loongson)). > > B) If rebase key is included, apply to tip of the indicated branch. > If, because the branch has changed, the patch no longer applies, then > we can report an apply failure. Then, submitter has to refactor their > patch and resubmit. That makes sense to me. > In either case, report the new results with an updated test result in > the email (i.e. report "_Testing PASS RETEST #1" instead of "_Testing > PASS" in the email body). Ack - makes sense here too. ^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: Email based retest request process: proposal for new pull/re-apply feature 2024-03-19 17:30 ` Patrick Robb 2024-03-19 17:53 ` Aaron Conole @ 2024-03-20 1:35 ` zhoumin 1 sibling, 0 replies; 11+ messages in thread From: zhoumin @ 2024-03-20 1:35 UTC (permalink / raw) To: Patrick Robb; +Cc: Adam Hassick, Aaron Conole, ci, dev On Tue, Mar 19, 2024 at 5:30PM, Patrick Robb wrote: > On Tue, Mar 19, 2024 at 4:37 AM zhoumin <zhoumin@loongson.cn> wrote: >> >> One more thing I want to confirm is whether we should apply the patch >> onto the branch commit which existed at the time when that patch was >> submitted or onto the latest tip of branch if users request doing >> rebase. Users probably request a recheck with `rebase` when the CI lab >> chose a wrong branch onto which apply the patch. I worry we may >> encounter conflicts when apply the patch onto the latest commit of the >> target branch if that branch is just updated before the request. >> >> > That's a good edge case to think about... but I also think if the > patch no longer applies cleanly on tip of intended branch, then we > would be correct to report an apply failure there. And then the > submitter should refactor their patch so it applies, and submit again. Yes, it is more reasonable for submitter. > So I think the process is like > > A) If retest is requested without rebase key, then retest "original" > dpdk artifact (either by re-using the existing tarball (unh lab) or > tracking the commit from submit time and re-applying onto dpdk at that > commit (loongson)). > > B) If rebase key is included, apply to tip of the indicated branch. > If, because the branch has changed, the patch no longer applies, then > we can report an apply failure. Then, submitter has to refactor their > patch and resubmit. Thanks for making the applying process more clear. > In either case, report the new results with an updated test result in > the email (i.e. report "_Testing PASS RETEST #1" instead of "_Testing > PASS" in the email body). Yes, I agree with this approach and reporting a new title for the retest result is necessary. ^ permalink raw reply [flat|nested] 11+ messages in thread
end of thread, other threads:[~2024-03-20 1:36 UTC | newest] Thread overview: 11+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2024-02-20 15:21 Email based retest request process: proposal for new pull/re-apply feature Patrick Robb 2024-02-20 18:12 ` Aaron Conole 2024-02-20 18:24 ` Patrick Robb 2024-03-01 14:36 ` zhoumin 2024-03-04 15:21 ` Aaron Conole 2024-03-07 17:06 ` Adam Hassick 2024-03-18 15:59 ` Patrick Robb 2024-03-19 8:36 ` zhoumin 2024-03-19 17:30 ` Patrick Robb 2024-03-19 17:53 ` Aaron Conole 2024-03-20 1:35 ` zhoumin
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).