Re: [Monotone-devel] Sync Issue

[aecooper]

Hello Zack,

The server does resume normal operation after the failing client disconnects.

On both occasions the client seemed to lock in the transmission stage just
after it uploaded a revision and had yet to download outstanding
revisions. On the first occasion I rebuilt his database and resubmitted
his revision. In the second case I tried a push and a pull which seemed to
correct the problem.

It doesn't occur that often (twice in 10 months). I will try fiddling
around with the 100s of revisions to download scenario and 1 to upload.

It isn't a big issue, this was really more a `have you seen this before'
question.

What I will do is try and make it repeatable with a script generated
database, see if it happens with the latest version or mtn and then if it
does submit a bug report with the details/test scripts.

Many thanks

Tony.
> On Sat, Mar 1, 2008 at 6:58 PM, Anthony Edward Cooper
> <address@hidden> wrote:
>>     We use monotone 0.35 at work and have been having some minor issues
>>  when syncing databases. The problem seems to happen when a database has
>>  a couple of revisions to upload and several hundred revisions to
>>  download (at a minimum). The sync sometimes locks up the client and the
>>  server locks out other clients while the offending client is left
>>  running. The server is idle waiting on poll, presumably waiting for
>> some
>>  subtransaction to complete. It resumes normal operation once the faulty
>>  client is terminated. Once the client has done this once it does it
>>  again when rerun.
>
> This does not sound like a known bug.  --Perhaps I should hedge that
> and say that we do know that a client that stops sending data at the
> wrong time can DoS the server.  What we haven't seen before is a
> client that does that absent a network or power failure.
>
> Does the freeze happen during the refinement phase ("finding items to
> synchronize") or the
> transmission phase (after the "bytes in | bytes out | revs in | revs
> out" banner appears)?
>
> Can you construct a test case (not involving 1GB of your company's
> data) that reliably reproduces the effect?
>
> zw
>