Rob Herring <[email protected]> wrote:
> On Thu, Jun 30, 2022 at 2:55 AM Eric Wong <[email protected]> wrote:
> >
> > Rob Herring <[email protected]> wrote:
> > > On Wed, Jun 29, 2022 at 11:27 AM Eric Wong <[email protected]> wrote:
> > > > Rob Herring <[email protected]> wrote:
> > > > > On Wed, Jun 29, 2022 at 10:30 AM Eric Wong <[email protected]> wrote:
> > > > > > Rob Herring <[email protected]> wrote:
> > > > > > > Hi,
> > > > > > >
> > > > > > > I'm using lei with lore where I have 2 queries which overlap. 
> > > > > > > Really,
> > > > > > > one is a subset of the other. On those overlapping threads, I'm
> > > > > > > finding that sometimes new messages are written to one mailbox 
> > > > > > > and not
> > > > > > > the other. (At least sometimes, the messages may be missing from 
> > > > > > > all
> > > > > > > mailboxes sometimes too. I'm not certain.) Using 
> > > > > > > --remote-fudge-time
> > > > > > > to force refetching seems to get the missing mails. I haven't 
> > > > > > > found
> > > > > > > anything strange in timestamps of the missing mails, but 
> > > > > > > otherwise am
> > > > > > > not sure how to debug this further. The queries are retrieving 
> > > > > > > full
> > > > > > > threads and the missing mails are in the threads, but not direct
> > > > > > > matches to the queries. I realize that's not a lot of detail to 
> > > > > > > go on.
> > > > > > > Suggestions on debugging this further?
> > > > > >
> > > > > > Is this with 1.8 or 1.7?
> > > > >
> > > > > Commit 68b53c888911 actually. So post 1.8.
> > > >
> > > > OK, thanks for that info.
> > > >
> > > > > > I forgot to note in the release notes, but there were some
> > > > > > SQLite usage-related fixes which could avoid missing messages.
> > > > > >
> > > > > > You'll need "lei daemon-kill" after upgrading to 1.8 to ensure
> > > > > > the new code is running.
> > > > >
> > > > > It's possible I haven't done that since updating though I do vaguely
> > > > > recall seeing something about needing to do that. Is there any way to
> > > > > tell before I restart it?
> > > >
> > > > Not really, but it's pretty cheap to restart (assuming there's no
> > > > long-running jobs).
> > >
> > > I've restarted and just hit this again.
> >
> > Ugh, sorry to hear that :<
> >
> > > > > > What might be interesting is to use the URLs lei prints and
> > > > > > comparing the results w/o lei.
> > >
> > > $ lei up --all
> > > # updating /home/rob/Mail/from-me
> > > # updating /home/rob/Mail/missing-cc
> > > # updating /home/rob/Mail/my-patches
> > > # updating /home/rob/Mail/pci
> > > # https://lore.kernel.org/all/ limiting to 2022-06-27 12:42 -0600 and 
> > > newer
> > > # https://lore.kernel.org/all/ limiting to 2022-06-27  9:50 -0600 and 
> > > newer
> > > # https://lore.kernel.org/all/ limiting to 2022-06-27 12:42 -0600 and 
> > > newer
> > > # /usr/bin/curl -Sf -s -d ''
> > > https://lore.kernel.org/all/?x=m&t=1&q=(dt%3A20220529211430..+AND+(f%3Arobh%40kernel.org+OR+f%3Arobh%2Bdt%40kernel.org))+AND+dt%3A20220627184226..
> > > # /home/rob/.local/share/lei/store 144/144
> > > # /home/rob/.local/share/lei/store 3/3
> > > # /usr/bin/curl -Sf -s -d ''
> > > https://lore.kernel.org/all/?x=m&t=1&q=((dfn%3Adrivers+OR+dfn%3Aarch+OR+dfn%3ADocumentation%2F*+OR+dfn%3Ainclude+OR+dfn%3Ascripts)+AND+f%3Arobh%40kernel.org+AND+rt%3A1640812470..)+AND+dt%3A20220627155025..
> > > # /usr/bin/curl -Sf -s -d ''
> > > https://lore.kernel.org/all/?x=m&t=1&q=(l%3Alinux-pci+dfn%3Adrivers%2Fpci%2Fcontroller+dt%3A20220529211430..)+AND+dt%3A20220627184226..
> > > # /home/rob/.local/share/lei/store 0/0
> > > # /home/rob/.local/share/lei/store 362/362
> > > # 0 written to /home/rob/Mail/missing-cc/ (0 matches)
> > > # https://lore.kernel.org/all/ 72/72
> > > # https://lore.kernel.org/all/ 4/4
> > > # https://lore.kernel.org/all/ 131/?
> > > # https://lore.kernel.org/all/ 184/?
> > > # https://lore.kernel.org/all/ 412/?
> > > # https://lore.kernel.org/all/ 603/?
> > > # https://lore.kernel.org/all/ 853/?
> > > # https://lore.kernel.org/all/ 1069/?
> > > # https://lore.kernel.org/all/ 1442/?
> > > # https://lore.kernel.org/all/ 1443/1443
> > > # 1 written to /home/rob/Mail/pci/ (75 matches)
> > > # 2 written to /home/rob/Mail/my-patches/ (148 matches)
> > > # 7 written to /home/rob/Mail/from-me/ (1805 matches)
> > >
> > >
> > > What I expected was 3 messages written to 'my-patches'.
> > >
> > > I think the problem is just simply that the new message missing
> > > doesn't match the query, but is a reply to a match. So with a date
> > > after the original match in the thread won't pick up anything. The 2nd
> > > URL above indeed only has 2 results. I guess I just have to fetch a
> > > wider window like a month every time? What's needed is a get any new
> > > messages in existing threads. I don't suppose there's an efficient way
> > > to do that?
> >
> > No, I don't think so.  I think this is a separate issue in lei...
> > "t=1" in the remote query expands threads in a time-agnostic
> > way, so I don't think that's the problem (though I may be wrong...).
> 
> Based on what the web interface presents, it sure seems like 't=1' is
> independent of the query. The results listed are only those that match
> the query and date range on the match.

Actually, for the HTML results, t=1 is ignored right now...
it's only for mboxrd downloads (via POST) at the moment...
That should be clarified/changed.

> For example, this query returns 3 matches:
> 
> https://lore.kernel.org/all/?x=m&t=1&q=((dfn%3Adrivers+OR+dfn%3Aarch+OR+dfn%3ADocumentation%2F*+OR+dfn%3Ainclude+OR+dfn%3Ascripts)+AND+f%3Arobh%40kernel.org+AND+rt%3A1641934905..)+AND+dt%3A20220630203819..
> 
> If I change 'dt' to 1 day earlier, I get 1 more match:
> 
> https://lore.kernel.org/all/?x=m&t=1&q=((dfn%3Adrivers+OR+dfn%3Aarch+OR+dfn%3ADocumentation%2F*+OR+dfn%3Ainclude+OR+dfn%3Ascripts)+AND+f%3Arobh%40kernel.org+AND+rt%3A1641934905..)+AND+dt%3A20220629203819..
> 
> That 4th match has a reply after 6/30, but the 1st query will not get
> the reply. This is all reproducible without lei involved at all.

Right.  t=1 only expands threads if they're linked via
References/In-Reply-To or (loosely) via matching Subjects.

> What seems to be needed is a 'thread date' which is the latest time
> for any message in a thread that matches. Or perhaps some way to
> separate the query from what's transferred. IOW, query for X, but only
> send results newer than some date.

Yeah, the thread ID info is stored independently of the search
index documents, though; so searching (Xapian) vs filtering
(SQLite) gets a bit tricky

I'll look into JMAP further before making more changes to the
data models to accomodate this.

Apologies for the delays; been stressed over other things :<

Reply via email to