Closed Bug 612233 Opened 14 years ago Closed 13 years ago

can not connect to server - tries to connect to old server after profile update

Categories

(Thunderbird :: Account Manager, defect)

Product:
Thunderbird
Component:
Account Manager
Platform:
x86_64
Linux
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

(Not tracked)

Status:
RESOLVED INCOMPLETE

People

(Reporter: doublehp, Unassigned)

Details

(Whiteboard: [needs protocol log])

User-Agent: Mozilla/5.0 (X11; DHP; Linux x86_64; en-US; rv:1.9.2.12) Gecko/20101030 Gentoo Firefox/3.6.12 Build Identifier: Mozilla/5.0 (X11; U; Linux x86_64; fy-NL; rv:1.8.1.23) Gecko/20100130 Lightning/0.9 Thunderbird/2.0.0.23 Mnenhy/0.7.5.0 In 2004-2005 i was living in UK. I subscribed to Pipex.net for ADSL services. I created a newsgroup account called "nntp" with nntp.pipex.net as server. In mid 2005 I moved to France, and subscribed to Free.fr, and updated my profile, and changed the server name to news.free.fr The bug happens almost always when i subscribe to groups, and, almost never in any other case (normal daily use). Bug happens since years. Problem is, especially if i subscribe several groups at the same time: "can not connect to server nntp.pipex.net". The only place where i can see this name today is the folder name, in the profile description: "newsrc file" and "Local directory". Outgoing SMTP by Pipex have been removed. Thunderbird can, at that time, connect to Free, list the group, and show me the new/recent subjects of the group, and can even sometimes show me message content even while it's still downloading headers ... but, at some point, I almost always end up with this popup. Pop-up which, due to various other bugs, blocks compleetely the interface. See bug #271587 (and it loves hiding under Firefox, so that i can't see it, and i get angry against frozen TB !!! ). So, why does it even try to connect to a server that is not pointed by my profile ? Reproducible: Sometimes Steps to Reproduce: No clue how to make other people repro this. But main lines: - create an nntp account pointing to service A - change server name to an other service B - subscribe some groups - try to browse the group during download of headers - get popup "can't connect to A" Reboots and TB-restarts may affect.
(In reply to comment #0) > and updated my profile, and changed the server name to news.free.fr > The only place where i can see this name today is the folder name, in the profile description: "newsrc file" and "Local directory". There are some places where old/initial server name is held, in addition to directory/file name. (a) prefs.js entry. Check via Config Editor(Tools/Options/Advanced/General) > mail.server.serverN.hostname=nntp.pipex.net When you changed server name, next entry was generated, but .hostname entry was kept for internal consistency. > mail.server.serverN.realhostname=news.free.fr Initially, .hostname was used. To support server name change, .realhostname was introduced, and .realhostname is used for server connection. (b) hostinfo.dat under ...\News\nntp.pipex.net directry. hostinfo.dat for news.mozilla.org in my environment. > version=1 > news.mozilla.org > lastgroupdate=1252375377 > firstnewdate=34173493 > uniqueid=0 > > begingroups > netscape.public.mozilla.performance.size-matters,,1,0,0 > (newsgroup names follow) I think one of aboves is cause, if problem upon subscribe only, and if any news group accessed at nntp.pipex.net exists at news.free.fr too. User usually doesn't change news server name. User usually creates news account if different news server is needed. This is because what newsgroups are supported at a news server depends on ISP's policy, and accessible newsgrops are usually different if different news server. I think it's reason why you are first problem reporter, if ".hostname != .realhostname" or "old/initial server name in hostinfo.dat" is cause of your problem. My guess is "old/initial server name in hostinfo.dat", because ".hostname != .relhostname" doesn't produce such kind of problem on POP3/IMAP account and hostinfo.dat is News only file. Can you check with news.free.fr in hostinfo.dat? Please keep backup of hostinfo.dat before test.
I couldn't see phenomenon you stated in comment #0, by changing news server from news.mozilla.org to news.opera.com, with Tb 3.1.5 on Win XP-SP2. I could see next only. (1) As newsgroups of news.mozilla.org doesn't exist at news.opera.com, Tb issued dialog like next for newsgropus subscribed at news.mozilla.org. a-news-group-of-mozilla is not found on NEWS.MOZILLA.ORG, unsubscribe? News server name in this message looks string in hostinfo.dat. (2) After unsubscribe by OK to dialog for all non-existent newsgroup, subscribe, unsubscribe etc. was normally executed for news.opera.com. (3) If dialog is kept for a while and replied "Cancel", alert of "connection timeout for news.opera.com" was shown. It looks for me Tb 3.1.5 works as designed for .realhostname on News too, although "use of string in hostinfo.dat for message" can be called Tb's bug and phenomenon (3) is not so well for user. Do you enable "Check for new mesages upon startup/every NN minutes"? You changed news server name setting and you still keep Tb running(no shutdown of Tb)?
> newsgroups of news.mozilla.org doesn't exist at news.opera.com wrong test: opera and mozilla are both reachable; nntp.pipex.net is not pingable at all for me from France. If you try to repro, replace the server by news.invalid.foobar ... > Do you enable "Check for new mesages upon startup/every NN minutes"? Yes ... i don't see the point. Every 5mn. I changed news server name only ONCE, in 2005 (why reconfigure a compleet profile, when you only need to modify one server field ???). Since that day, I have unsub, and re-subscribe to many groups. In fact, i have unsubscribed to all group I was reading before the change. I shutdown computer at night. It's hard to test, because I can't always repro the bug. Just now, i have not changed the conf yet; i see newsrcname=nntp.pipex.net in ./News/nntp.pipex.net/hostinfo.dat . I subscribed to some random group, and clicked on in in the pannel; it started downloading headers. Then i clicked an other group in the pannel, back to the new one, and it started download over (count was at 12000/48000, and restarted at 1000/48000); three times in one minute, each time i clicked on an other group (stressing TB usually helps bugging :D ). I could click on one message to view it; not, TB is locked up: it was around 27000/48000, not using network connexion (0B/s), but 100% CPU (that's strange, is TB still monothreaded ? I have quad core, 400% available, TB only uses 100% when it locks-up). Windows not content not refreshed; when process is not responding, E17 makes the decoration blink; it's the way for E17 to inform user that window content is not relevant, and not fresh. Any way, after 10mn, window stopped blinking, and i see 191333 messages. Around 4 times 48000; probably because of the count restart. Several messages can be seen 4 times; not all. Any way, for what matters to us today, I did not have the error pop-up. THIS TIME, I did not have it. I says in #0 "almost always", not always. I start again with an other group. This time, i just subscribe, and don't click on it. 5mn later, no issue. Don't know how to repro. This morning, it works.
Before changing anything in the conf, I want to have a known to bug procedure; as long i I don't have 100% reproductible procedure, i can't test anything.
Arg, and now ... "The group whatever does not appear to exist on the host nntp.pipex.net. Would you like to unsubscribe from it ?"
(In reply to comment #4) > Before changing anything in the conf, I want to have a known to bug procedure; > as long i I don't have 100% reproductible procedure, i can't test anything. actually, if it's hard for you to reproduce, then it's probably even harder or impossible for other people. wada is giving you good guidance.
Dominique, could you use a tool like wireshark and verify that TB is really trying to access the old server ?
Note: before I have the timeout pop-up, I have *ALWAYS* seen "trying to connect to pipex.net" in the status bar (bottom left of window). I mean, before failure (failure is because nntp.pipex.net is reachable only for Pipex customers), TB warns it's trying to connect to the "wrong server". So, the main point of repro, as stated in the subject, is not to get a "can not reach host", but in "trying to reach old host, even after profile update". If you changed server A into B, and see "trying to reach A" in the bottom left of window, then, you have my bug. The time-out error is an expected side effect of the bug; not the main point. You don't need to wait for that error; i don't mind the error itself. I mind about the trying wrong host. Generally speaking, it means that when user rectifies his conf, TB does not car about the changes. It means there is a fault in the design of the account manager. Yes, I admitt it's hard to repro; it's not all the time; it's usually just after new group registration. Can't say more. The bug is confirmed already by Wada: he is right: i can see pipex.net in hostinfo.dat . Reproduction of bug depends on ... if TB uses this file or not (or THAT line in the file). If some dev can tell me when this line/file is used, then, we will be able to explain a 100% reproduction method. As you tell me when/how/why TB uses this line, we got a stable repro. You already know how to fix: Wada already explained the difference between account creation, and account update. As reporter and "stupid user", i have nothing more to teach you about it. You already know and understand it better than me. And i desagree about "usually people just delete and recreate account". Example: when you initially create the profile, you make a typo, a mistake when typing the server. Then you fix the typo ... TB will keep (some times) trying to acces the wrong name. You can *NOT* expect user to always delete and re-create account for every single typo. Good luck.
(In reply to comment #3) > > newsgroups of news.mozilla.org doesn't exist at news.opera.com > > wrong test: opera and mozilla are both reachable; nntp.pipex.net is not > pingable at all for me from France. If you try to repro, replace the server by > news.invalid.foobar ... > I changed news server name only ONCE, in 2005 I emulated your settings and operations. Quick check results. (1) New profile, news account of news.php.net, subscribe some newsgroups, check for new message upon startup/every NN minutes, some restarts of Tb and newsgroup accesses. (2) Change news server name from news.php.net to news.mozilla.org, terminate Tb (3) Edit C:\WINDOWS\system32\drivers\etc\hosts 127.0.0.1 news.php.net ping news.php.net => timeout, as expected ping news.mozilla.org => OK, as expected (4) Restart Tb, with NSPR logging SET NSPR_LOG_MODULES=nntp:5,nsHostResolver:5 (5) Tb requested DNS query for news.mozilla.org in .realhostname as expected, by "check for new message upon startup". But NNTP log was not seen in log. (6) Click newsgroup subscribed from news.php.net. => Error message of "Connection timeout for NEWS.MOZILLA.ORG". NNTP request was not seen in log. Subscribe => subscription list is not refreshed NNTP request was not seen in log. => Unable to subscribe newsgroup on news.mozilla.org (7) Uncheck "check for new message upon startup/every NN miute", restart Tb (8) Click newsgroup of news.php.net. => "The group whatever does not appear to exist on the host NEWS.PHP.NET. Would you like to unsubscribe from it ?" Reply OK for all non^existent newsgroup. All non-existent newsgroups are removed. Subscribe => Subscription list of news.mozilla.org is shown. Subscribe/unsubscribe is normally executed. I changed news.php.net in hostinfo.dat to news.mozilla.org manually during test once, but it was automatically changed back to news.php.net by Tb. String of .hostname looked written in hostino.dat always. As manual change is involved in above tests, some test results may not be accurate, but many of above tests were executed with "news.php.net in hostinfo.dat". During test, I couldn't see NSPR log of DNS lookup for old news.php.net. Problem looks for me: Old news server name(nntp.pipex.net in your case) in message comes from hosinfo.dat, which is string set in .hostname. Connection to nntp.pipex.net(in your case) is not requested by Tb because server specified in .realhostname is news.free.fr(in your case). Tb wrongly waits for connection to server specified in hostinfo.dat, if a newsgroup is non-existent and is accessed by "check for new messages ...". So, if .realhostname != .hostname(==server name in hostinfo.dat), and if server specified in .hostname is not reachable or accessble, timeout error or connection error for Tb happens. Can you get NSPR log for nsHostResolver:5? If possible, check DNS lookup request for old nntp.pipex.net in NSPR log when problem happens, please.
A part of NSPR log for succesful newsgroup access. > [5576] 0[192c140]: (4733bc0) fullPath = /mozilla.evangelism > [5576] 0[192c140]: (4733bc0) opening connection to news.php.net on port 119 > [5576] 0[192c140]: (4733bc0) setting busy to 1 > [5576] 0[192c140]: (4733bc0) ParseURL > [5576] 3992[192c280]: nsHostResolver::ResolveHost [host=news.mozilla.org] Request to nsHostResolver is host=news.mozilla.org, so connection is requested news.mozilla.org as expected. However, next is also seen in log. > opening connection to news.php.net on port 119 I don't know Tb is really waiting for connection to news.php.net, or waiting for connection to "server of .realhostname for which .hostname==news.php.net(==string in hostinfo.dat) is set".
For news.php.net(old news server) in next message and NSPR log. > php.announce does not appear to exist on the host NEWS.PHP.NET. Would you like to unsubscribe from it ?" > opening connection to NEWS.PHP.NET on port 119 I restarted Tb after change news.php.net in hostinfo.dat to news.php.neX. Server name in message/log was still news.php.net, and string in hostinfo.dat was not altered by Tb until next subscribe operation. So, news.php.net comes from .hostname of prefs.js which is used in internal URL for news server or newsgroup. It was not from hostinfo.dat. Server name data in hostinfo.dat was merely copied from .hostname after subscribe. Sorry for my confusion and wrong guess. And, I couldn't observe funny phenomena such as "subscribe shows nothing" any more. In any test, "php.xxx does not appear ...host NEWS.PHP.NET. ... unsubscribe from it ?" appeared upon any click of newsgroup of php.xxx. Funny phenomena I saw was perhaps result of my setup error in previous test. In my evironment, Tb 3.1.5 works as designed for .realhostname on NNTP too. Problem in my environment/in my test is only confusing string which comes from .hostname instead of .realhostname in some messages, even after implementation change by ".realhostname support".
(In reply to comment #0) > The bug happens almost always when i subscribe to groups, and, almost never in > any other case (normal daily use). Bug happens since years. > > Problem is, especially if i subscribe several groups at the same time: "can not connect to server nntp.pipex.net". I couldn't see connection request to old news server, and I couldn't see error message like "can not connect to server <old_news_server>" in my test. I saw "Connection timeout to <new_news_server>"(alert message) and "Failed to connect to <new_news_server>"(Activity Manager) in my test. I could see <old_news_server>(NEWS.PHP.NET in my test) in next message only. > "php.xxx does not appear ...host NEWS.PHP.NET. ... unsubscribe from it ?" At where was the "can not connect to server nntp.pipex.net(==<old_news_server>)" shown? At where did you request subscribe of newsgroups? (a) Subscribe/"Current News Groups" tab? (b) Subscribe/"New Groups" tab? As no new newsgroup was added at server during my test, I couldn't check (b). Anyway, please check which is your problem. (i) Tb really requested connection to <old_news_server>(string of .hostname), and it failed.　 (ii) Server name in error message is string of .hostname instead of string of .realhostname.
I now have reproductible problem: my actual NNTP server is still free.fr (and old was Pipex). I have subscribed to some group yesterday; after reboot, I now have this message every 5 or 10mn: The newsgroup fr.rec.cinema.selection does not appear to exist on the host nntp.pipex.net. Would you like to unsubscribe from it? Either the group is available on Free.fr, and the profile is upside down, or, it does not, and the message is wrong. To get this, i did this yesterday: - manage subs - tick 5 new groups - OK - click on first group - VERY RAPIDLY, click on this one, which was second new one in my list (within 0.2s after clicking the first group: TB was still negociating "connexion" with server). Doing things fast has impact on this bug. So, if the group does not exist anymore, either it was removed in the 10s between my listing server, and trying to access it, or, i could finaly make TB memorise the bug, and it now persists over reboots and restarts.
did you ever considered removing the newrc file from your folder ? And you should probably clean up you prefs.js with the old account information.
And why not format my hard drive, and install Windows 95 ? So, Mozilla products keep on working only if we purge our profiles on every single update ? No, I will NOT reconfigure my softwares on every single update. The dev have to manage to make the products compliant with older profiles, and import them properly.
plop, it seems no one currently subscribed can reproduce with your steps. do we need a log or newsrc file?
Whiteboard: [closeme 2011-11-01][needs protocol log]
Probably because in modern society, I may be the last human using a 12 years old computer profile; every one just throught things away, a don't mind any more about data loss. When things are broken, just put them in the bin, flash, flush, erase, format ... and start over. Mozilla dev team already proved me they also work this way: if things are broken after a version update, and work after profile wipe, then, they consider there is no bug. I will NOT recreate all my email accounts and filters and rules and scripts after each software update. The product must be able to stand updates by itself.
(In reply to plop from comment #17) > I will NOT recreate all my email accounts and filters and rules and scripts > after each software update. The product must be able to stand updates by > itself. plop, thanks for all the info. If it is easy for you to reproduce, then making a small test profile [1] with just for the news account is sufficient. Then we will know whether the problem is related to your profile, and perhaps help us understand if anyone else should be able to reproduce it. If if it reproduces, an nntp log will be helpful [1] https://support.mozillamessaging.com/en-US/kb/Using+Multiple+Profiles
Whiteboard: [closeme 2011-11-01][needs protocol log] → [closeme 2011-11-25][needs protocol log]
Status: UNCONFIRMED → RESOLVED
Closed: 13 years ago
Resolution: --- → INCOMPLETE
Whiteboard: [closeme 2011-11-25][needs protocol log] → [needs protocol log]
You need to log in before you can comment on or make changes to this bug.