On Sat, Jun 2, 2018 at 12:20 PM, Martin Perina <mperina(a)redhat.com&gt; wrote: > Hi, > > recently [1], [2] were raised, but the main reason for those bugs was the > we are no paying enough attention to consequences when doing backports. > Below are two examples which we need to pay more attention to: > > 1. Database upgrade scripts naming and numbering > - there's no problem if the same script has different name or number > between y-streams (for example 4.3 and 4.2), but names and number need to > match between z-streams > - if we are going to do async release (for example 4.2.3.z) and we > need to backport to async branch patch which includes db upgrade script, we > need to make sure to align db scripts in branch from which next z-stream > will be built (for example 4.2, patch [3] can be used as an example how to > fix this issue) > > 2. Use the same name of database upgrade script when doing backports > - when backporting some patches to z-stream branch please use the same > db script name and change only db script number as needed, otherwise > backports are quite confusing (for example [4] and [5]) > > > Patch [3] fixes the situation, so when included into 4.2.4 build it > should allow smooth upgrade from 4.2.3.z to 4.2.4, but users which already > performed upgrade are in trouble. @Eli is there any way how we can force > execution of db scripts, which were skipped due to the mess in names > between 4.2.3.z and 4.2.4 to help users which have already broken database? > > > @Tal/@Piotr, could you please pay even more attention when merging > patches containing db scripts to z-stream async branches (normal branch, > for example ovirt-engine-4.2, should be fine, but ovirt-engine-4.2.3.z > requires more attention)? I agree that developer who backports the patch > should handle the situation (feel free to ask infra team if unsure about > consequences), but you are the last safety check we have. > It is a bit problematic from my side, let's say script number 0110 makes its way to 4.2 and has to be backported to 4.2.3.z but since two other patches which include db scripts were not for 4.2.3.z but only 4.2, in this way the script number has to be changed to 0090 in 4.2.3.z since the upgrade script does not allow gaps of more than 10 in script numbers

> Thanks a lot > > Martin > > [1] https://bugzilla.redhat.com/1583562 > [2] https://bugzilla.redhat.com/1583664 > [3] https://gerrit.ovirt.org/91874 > [4] https://gerrit.ovirt.org/90695 > [5] https://gerrit.ovirt.org/90698 > > > -- > Martin Perina > Associate Manager, Software Engineering > Red Hat Czech s.r.o. >

On Sat, Jun 2, 2018 at 12:20 PM, Martin Perina <mperina(a)redhat.com&gt; wrote: > Hi, > > recently [1], [2] were raised, but the main reason for those bugs was the > we are no paying enough attention to consequences when doing backports. > Below are two examples which we need to pay more attention to: > > 1. Database upgrade scripts naming and numbering > - there's no problem if the same script has different name or number > between y-streams (for example 4.3 and 4.2), but names and number need to > match between z-streams > - if we are going to do async release (for example 4.2.3.z) and we > need to backport to async branch patch which includes db upgrade script, we > need to make sure to align db scripts in branch from which next z-stream > will be built (for example 4.2, patch [3] can be used as an example how to > fix this issue) > > 2. Use the same name of database upgrade script when doing backports > - when backporting some patches to z-stream branch please use the same > db script name and change only db script number as needed, otherwise > backports are quite confusing (for example [4] and [5]) > Patch [3] fixes the situation, so when included into 4.2.4 build it > should allow smooth upgrade from 4.2.3.z to 4.2.4, but users which already > performed upgrade are in trouble. @Eli is there any way how we can force > execution of db scripts, which were skipped due to the mess in names > between 4.2.3.z and 4.2.4 to help users which have already broken database? > ​unfortunately, those should be run manually or merge all to one re-entrant script and add it as an additional upgrade script.