Courier Mail Server

Calendaring in SqWebMail

This is currently a work in progress. At this point, the calendaring implementation includes basic calendaring/scheduling resources.

Dogfood: "pcp" is the acronym for "Personal Calendaring Protocol".

Enabling calendaring

To enable calendaring, create the file calendarmode in the configuration directory (/usr/local/share/sqwebmail, or $sysconfdir, usually /usr/lib/courier/etc). The calendarmode file should contain a single word: either "local" - enable account calendaring; or "net" - enable calendaring in groupware mode. Example:

echo "local" >calendarmode

Additionally, if groupware calendaring is enabled, the pcpd daemon process must be started. This is now taken care of automatically by the sqwebmail startup script (sqwebmaild.rc script in the sqwebmail distribution, and the webmaild script in the Courier distribution). The next step is to set up individual access control lists, which are described below.

NOTE: if you've installed SqWebMail/Courier binary RPMs, the RPMs will automatically take care of starting pcpd at system boot time if groupware calendaring is enabled.
NOTE: check your umask, calendarmode needs to be globally readable.

Technical implementation

There are two calendaring modes:

Local mode

Local calendar access requires very little additional resources, in addition to basic webmail support, and it needs very little overhead. Calendaring information is saved in $HOME/Maildir/calendar, which is automatically created during first webmail login after enabling calendaring.

Groupware mode

Groupware mode offer the ability to synchronize events on multiple calendar. Groupware mode requires additional server resources. A rule of thumb is that groupware mode will require between 10% and 100% of CPU and I/O bandwidth, when compared to basic webmail. Groupware may also require additional disk space, since copies of the event data is copied to each calendar. Events may have large attachments, so allow for plenty of disk space.

In all cases, calendaring is designed to be used in private, non-public settings. This calendaring implementation will not work well in a public webmail environment. Currently, quotas are not implemented on the calendar data. SqWebMail does not work with filesystem quotas. SqWebMail creates cached folder indexes, and an inability to create a folder index will result in a total application failure. SqWebMail does support a software quota implementation, however the software-based quota currently does not include calendaring data. This issue may be addressed in the future.

This calendaring implementation is built on top of an API protocol library that is designed to be flexible and extensible. Only basic calendaring is implemented right now. The API library uses locks when accessing calendar data. Locks are not really necessary in local mode, but are needed for groupware mode. The API library uses dot-lock files, which are designed to work on a network filesystem, as long as the clocks on the clients and servers are synchronized. When a network filesystem is used it is essential to make sure that the clocks on all machines are synchronized with NTP, or an equivalent clock synchronization protocol.

Overview

Basic calendaring allows creation and modification of calendar events. A calendar event is handled similarly to mail messages. A calendar event has a subject, and a description. It is possible to attach files to calendar events, similar to how files can be attached to messages. This is mainly useful in groupware mode. Note however that group events place individual copies of the event on each calendar, and large attachments can quickly eat up storage space.

Individual events may be marked as cancelled, or completely removed from the calendar. A cancelled event still appears on the calendar, but new events will not check events marked as cancelled for scheduling conflicts. Adding a new event checks only non-cancelled events for any scheduling conflicts.

Events are created and edited like regular E-mail messages. Until an event is added to the calendar, it is saved in the Drafts folder while it is being edited. To cancel an event without saving it on the calendar, open the Drafts folder, and delete it.

Groupware Calendaring

NOTE: Groupware calendaring is experimental at this time. If you're using groupware calendaring on platforms that PAM for authentication, be sure to configure the PAM library to authenticate the "calendar" service.

It is helpful to know how groupware mode is implemented, in order not to be surprised by some of its quirks. The details of initializing groupware mode have already been covered, and are not repeated here.

Groupware mode works essentially by replicating calendar events across multiple calendaring. Whoever creates a calendar event specifies additional "participants". When groupware calendaring is enabled, additional fields on the New Event screen are used to enter additional participants' E-mail address. All participants' mailboxes must reside on the same server. In far distant future it will be possible to cluster pcpd across multiple servers. When the event is saved, the pcpd daemon replicates the event to all participants' calendars. Subsequent modifications to the event are similarly replicated.

But, before all this happens, it is necessary to set up access control lists. Access control lists determine who is authorized to place events on whose calendars. You wouldn't want everyone, and their aunt, to place random events on your calendar, do you?

The preferences screen will have a link on the bottom to a separate preferences page where access controls are entered. To allow someone to place events on your calendar, add their E-mail address to your access control list. Right now, two access controls are available:

Everyone is responsible for setting up their own access control lists. Right now, the "MODIFY" access control is always selected, because that's the only thing that makes sense. There will be other access controls added, in some distance future...

Updating groupware events

Only the original person who created an event is capable of modifying it. Other event participants may also mark the event as cancelled on their calendar, but should not make any other changes. Remember that the event creator is responsible for propagating event updates. A participant can still make changes to that event, but those changes will not be replicated, because only the event on the creator's calendar has the necessary internal replication linkage. If a participant modifies a replicated event, if the original creator makes any subsequent updates, the replication process won't be able to automatically delete the original event on the participant's calendar, to be replaced with the updated event. The original replicated event is gone, it's been updated on the participant's calendar. The updated event looks like a completely separate event, which probably has the same starting/ending times, and will trigger a scheduling conflict.

Participants may mark replicated events as cancelled, which will not break the replication link. That's what cancelling/uncancelling was originally designed for. A participant may try to completely delete the replicated event, but it'll be back if the original event creator updates the event, which will cause it to be replicated back again out to everyone's calendar.

The original event creator may add/remove participants as part of updating the event. The event will be properly added/removed from the affected calendars, accordingly (and hopefully).

Suppose an event has been replicated to someone's calendar, and this participant then went ahead and removed the creator's access control list entry. If the original event creator attempts to save an update to the event, the update will fail to authorize. There's an additional option on the new/update event page called "Ignore errors updating other participants' calendars". That's self-explanatory - this error condition will be ignored, and the offender is automatically removed from this event's list of participants.

Entering E-mail addresses

As described previously, groupware events are created by entering the participant's E-mail addresses. For this document's purposes, an "E-mail address" means the login ID that's provided to log into SqWebMail. Someone may customize their From: header, but that doesn't count. It's their webmail login address that matters.

The E-mail address should include @domain. If absent, it will default to the creator's login E-mail domain. Therefore, in most cases it's not really necessary to explicitly tack on @domain when entering E-mail addresses, the default will be correct most of the time. If the default is wrong, it will be very obvious.

The calendaring screens automatically check the address book, when displaying events. If the E-mail address is found in the address book, the calendaring screens will quietly show the person's name, in addition to the E-mail address. Initialize those address books.

Issues

There are some minor design issues that may be addressed in the future. Some of them are already obvious, from the preceding documentation.

Event updates

An existing event is updated essentially by making a copy of it, and saving the updated event as a new event, while simultaneously deleting the old event. It's possible to open an existing event, for editing, more than once. If one copy of the event is placed back to the calendar, the old event information is removed, and the new event information takes its place. If another, previously opened, copy of the old event is opened from the Drafts folder, edited and saved, the updated event already on the calendar is NOT removed, and essentially another event is added to the calendar (assuming no scheduling conflicts with the previously updated copy of the event).

This should not be a major issue, since a scheduling conflict will be reported in most cases, thus bringing this issue to light. Note that it's always possible to manually override any scheduling conflicts, and place conflicting events on the calendar, but a manual override is always required. It's also possible to detect an attempt to save a duplicate updated, because the old event is no longer on the calendar, however this will raise other issues concerning groupware mode. The current approach is the simplest way to implement editing of existing events, and any complications can always be easily fixed by hand.

Expired calendar events

Calendar events are automatically deleted 30 days after their last occurence.

Recurring events

When asked to generate weekly/monthly/annual recurring dates for an event, until a certain date, the system stops after generating 100 dates. This is to prevent bad input from generating a huge list.

Limitations

On most systems, calendaring works only until year 2038, when the 32 bit time counter wraps around. On 64-bit systems calendaring will work beyond year 2038.

Localization

Calendaring get the date/time format, names of the days of the week, and names of the months of the year, from the locale settings. The calendar's locale can be set by initializing html/en-us/LOCALE (in the stock tarball) which is installed either in /usr/local/share/sqwebmail or /usr/lib/courier/share/sqwebmail by default. Note that the LOCALE setting will be reflected in all other SqWebMail functions. If the locale text uses a character set other than iso-8859-1 then it will also be necessary to initialize the CHARSET file.

If the locale is changed, the sample dates in html/en-us/newevent.html may also need changed to reflect the locale's weekday and month names. Date entry in calendaring uses the locale's weekday and month names as well. The strings 'tomorrow' and 'today' are also set in the newevent.html file - look inside the hidden comments at the beginning of this file.

NOTE: SqWebMail accepts dates either in MM/DD/YYYY or YYYY.MM.DD no matter what locale is in effect. This is for ENTERING dates only - the dates are always displayed according to the current locale This will be fixed in a future release.
NOTE: Error messages generated by the internal calendaring protocol library are in English, and at this time there is no flexible mechanism to localize them. This will be fixed in a future release.