elconv deletes everything, posted by Konstantin Olchanski on Wed May 20 01:49:37 2015
|
Converting from elog 2.9.something to new elog 3.1.0 elogd refuses to start, instructs running elconv in one logbook.
When I do so, elconv converts a existing mhttpd-style elog entries to the new format (the corresponding new-format entries already exist)
and deletes everything else - this is very bad.
So there are 2 bugs:
- elogd should not tell us to run elconv when both old-style and corresponding new-style elog entries exist
- elconv should not delete all existing new-style elog entries.
I confirm that elconv *does* delete all new-style elog entries - with strace, I see it issue "unlink" on every elog entry.
What a disaster!
K.O. |
this elog errors sending email, posted by Konstantin Olchanski on Wed May 20 01:52:23 2015
|
this elog gives errors sending mail through PSI email server. (did not capture the error messages, sorry). K.O. |
edit somebody else's draft, posted by Konstantin Olchanski on Wed May 20 01:54:55 2015
|
this elog offers me to edit a draft message, then yells at me "only some other user can edit this draft!!!".
methinks I should only be offered to edit draft messages that I own or I can edit. K.O. |
elogd moves elog entries, posted by Konstantin Olchanski on Wed May 20 01:59:17 2015
|
elogd 3.1.0 moves all elog entries into year-named subdirectories. this feature makes it incompatible with older elogs and so should be clearly mentioned in the documentation,
in the release announcement and in the release and migration notes. K.O. |
Re: elogd moves elog entries, posted by Konstantin Olchanski on Wed May 20 20:03:06 2015
|
> Stefan told me that the change was because some users were having thousands of yymmdda.log files
> in the logbook directories
I am one of those users. The elog for the ALPHA experiment at CERN goes back to 2006 or so,
with large volume of messages and huge number of attachments. The MIDAS forum elog goes back to 2003.
The TRIUMF DAQ internal elog goes back to 2001.
I think the new organization is an improvement.
K.O. |
Re: elogd complains about unknown cookies, posted by Konstantin Olchanski on Sat May 23 02:49:22 2015
|
> > elogd is spewing these messages about unknown cookies:
> >
> > Received unknown cookie "is_returning"
> > Received unknown cookie "__utma"
> > Received unknown cookie "__utmz"
> > Received unknown cookie "SSESSee3cc9c70bedf9a840203765bf409d7b"
> > Received unknown cookie "SESSee3cc9c70bedf9a840203765bf409d7b"
> > Received unknown cookie "MidasWikiUserID"
> > Received unknown cookie "MidasWikiUserName"
> > Received unknown cookie "MidasWiki_session"
> >
> > K.O.
>
> Delete your cookies ;-)
>
No, go. The MidasWikiUserName cookie will log me out of the MidasWiki, etc.
But why are MidasWiki cookies sent to the elog? Ah, yes, they all run from the same https://midas.triumf.ca.
This explains why elog complains about unexpected cookies - elogd does not expect to receive
mediawiki cookies - it does not expect to share the https:// connection with somebody else.
This is good to know for the future. (I.e. if mediawiki is confused by elogd cookies).
K.O. |
Re: elogd moves elog entries, posted by Konstantin Olchanski on Sat May 23 02:53:31 2015
|
>
> "News for each version can be seen in the changelog (->http://midas.psi.ch/elog/download/ChangeLog)"
>
> And for version 3.0.0 it states:
> - Create one logbook subdirectory pear year
>
The text should read:
the first time you run the new elogd against old elog data,
all existing entries will be moved into by-year subdirectories
and the old elogd will not work anymore.
K.O. |
Cannot download large attachments, posted by Konstantin Olchanski on Tue Jul 14 19:29:17 2015
|
Older versions of elogd have a problem with sending long attachements - the send() syscall is not protected against interrupt by SIGALARM. This seems to be fixed in non-SSL builds of current elog (version 3+),
but for SSL builds, the error handling for the SSL_write() function looks incorrect. In addition, from OpenSSL documentation it is not clear if SSL_write() can handle signal interrupts at all.
K.O. |