index problems after update

[Hajo Locke]

Hello,


Am 21.02.2019 um 23:06 schrieb Adi Pircalabu via dovecot:
> On 2019-02-21 22:18, Sami Ketola via dovecot wrote:
>>> On 21 Feb 2019, at 12.23, Hajo Locke via dovecot
>>> <dovecot at dovecot.org> wrote:
>>> I think mbox+procmail is a classic setup and wide used and good
>>> solution for many usecases. Same setup we use many years.
>>> We run ~2 mio mailboxes. our automated systems depends on this
>>> setup. creating mailboxes, managing mailboxes, creating automated
>>> filterrules, backupsystem to tell something of them. we can not
>>> switch our whole mailsetup to work around this bug.
>>> How to get a dump if dovecot not crashing but has wrong behaviour? I
>>> would like to help and provide useful info, but it depends on kind
>>> of problem.
>>> I think if a classic setup is not working in dovecot any more, this
>>> is a serious problem.
>>
>> In you first email to this thread it says:
>>
>>> Feb  8 08:45:37 hostname dovecot[14882]: imap(myuser): Fatal:
>>> master: service(imap): child 14135 killed with signal 6 (core dumped)
>>
>> So imap is crashing and even dumping a core.
>>
>> Also I must disagree with your mbox+procmail statement. mbox has
>> always been very unoptimised mailbox format and everyone should be
>> emphasised not to use it.
>> Also that combination has always had problems with indexing and file
>> locking. I would not use it on high volume mailservers. Or even medium
>> volume mailservers.
>
> Not directly affected by this issue since I'm not using mbox for any
> production system nor have I for many years. And it'd take a lot of
> effort to convince me to use mbox for anything someone would even dare
> to classify, even remotely, as "production". But if I understand OP's
> point of view correctly, he's not arguing necessarily for or against a
> specific mailbox format. Instead, he's flagging a regression and
> people will be very reluctant to upgrade or even adopt a certain
> feature in a new release of a product if regressions are seen as
> acceptable. Something that previously worked in an otherwise unchanged
> environment stopped working after an upgrade and this is a regression.
> Trying to convince people to move away from mbox is a very sensible
> approach, I'm all for it, but in cases like this not practical.
>
yes, thanks for your support. I think this is the case here.

As conclusion:
We believe mdbox is better storage format, but we cant switch 2mio
mboxes + peripherie adhoc.
We did a downgrade to 2.2.26.0 a few weeks ago. 2.2.26.0 is first
version usable with openssl 1.1.0 and solves this issue. so this dovecot
problem must introduced after 2.2.26.0 and leads to a problem where
dovecot is not able to fulfill its basic purpose.
until this downgrade we had a lot of problems and complaints and felt
forsaken.
Too bad that there is no really interest to fix mbox related problems.

Thanks,
Hajo