+
+Usage
+-----
+
+In my `~/.xinitrc` I have something like the following:
+
+```sh
+( $BIN/khatus \
+ --wifi_interface 'wlp3s0' \
+| stdbuf -o L tee \
+ >(stdbuf -o L "$BIN"/khatus_bar \
+ -v Opt_Mpd_Song_Max_Chars=10 \
+ -v Opt_Net_Interfaces_To_Show=wlp3s0 \
+ -v Opt_Pulseaudio_Sink=0 \
+ | "$BIN"/khatus_actuate_status_bar_to_xsetroot_name \
+ ) \
+ >(stdbuf -o L "$BIN"/khatus_monitor_energy \
+ | "$BIN"/khatus_actuate_alert_to_notify_send \
+ ) \
+ >(stdbuf -o L "$BIN"/khatus_monitor_errors \
+ | "$BIN"/khatus_actuate_alert_to_notify_send \
+ ) \
+) \
+2> >($BIN/twrap.sh >> $HOME/var/log/khatus/main.log) \
+1> /dev/null \
+&
+```
+(where `twrap` is a simple script which prefixes a timestamp to each line)
+
+The idea is to support appending any number of ad-hoc, experimental monitors by
+giving maximum flexibility for what to do with the sensor outputs, while
+maintaining some uniformity of msg formats (again, to ease ad-hoc combinations
+(e.g. Does the CPU get hotter when MPD is playing Wu-Tang?)). `khatus_bar`,
+`khatus_monitor_energy` and `khatus_monitor_errors` are just some initial
+examples.
+
+Design
+------
+
+### 2.0
+
+In an effort to simplify the components and their interfaces, I removed the
+concept of a global controller from the previous design (which, at least for
+now, is superfluous), so now it is essentially a pub-sub - parallel publishers
+(sensors) write to a pipe, which is then copied to any number of interested
+subscribers that can filter-out what they need and then do whatever they want
+with the data. Status bar is one such subscriber:
+
+`P1 > pipe&; P2 > pipe&; ... PN > pipe&; tail -f pipe | tee >(S1) >(S2) ... >(SN) > /dev/null`
+
+The cool thing is that, because the pipe is always read (`tail -f ... > /dev/null`),
+the publishers are never blocked, so we get a live stream of events to which we
+can attach any number of interested subscribers (` ... tee ... `) and, because
+the pipe is named, if a subscriber needs to - it too can publish something to
+the pipe without being blocked.
+
+```
+parallel +----------+ +----------+ +----------+
+stateless | sensor_1 | | sensor_2 | ... | sensor_n |
+collectors +----------+ +----------+ +----------+
+ | | | |
+ data data data data
+ | | | |
+ V V V V
+multiplexing +-------------+-----------+---------+
+to a pipe |
+ |
+ V
+copying to +-------------+-+---------+---------+
+subscribers | | | |
+ V V V V
+ +------------+ ... +----------------+
+any number of | status bar | | energy monitor |
+parallel +------------+ +----------------+
+subscribers | |
+ V V
+ +----------------+ +-------------+
+ | xsetroot -name | | notify-send |
+ +----------------+ +-------------+
+```
+
+### 1.0
+
+This was an improvement of having everything in one script, but the controller
+was still way too complicated for no good reason.
+
+```
+parallel +----------+ +----------+ +----------+
+stateless | sensor_1 | | sensor_2 | ... | sensor_n |
+collectors +----------+ +----------+ +----------+
+ | | | |
+ data data data data
+ | | | |
+ V V V V
+serial +----------------------------------------------+
+stateful | controller |
+observer +----------------------------------------------+
+ |
+ decision messages
+decision |
+messages |
+copied to |
+any number |
+of interested |
+filter/actuator |
+combinations |
+ |
+ V
+ +-------------+-+---------+---------+
+ | | | |
+ V V V V
+parallel +------------+ +------------+ +------------+
+stateless | filter_1 | | filter_2 | ... | filter_n |
+filters +------------+ +------------+ +------------+
+ | | | |
+ V V V V
+parallel +------------+ +------------+ +------------+
+stateless | actuator_1 | | actuator_2 | ... | actuator_n |
+executors +------------+ +------------+ +------------+
+ | | | |
+ commands commands commands commands
+ | | | |
+ V V V V
+ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ ~~~~~~~~~~~~~ operating system ~~~~~~~~~~~~~~~~~
+ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+```
+
+### 0.x
+
+A single script, re-executed in a loop at some intervals, serially grabbing all
+the needed data and outputting a status bar string, then passed to `xsetroot -name`,
+while saving state in files (e.g. previous totals, to be converted to deltas).
+
+This actually worked surprisingly-OK, but had limitations:
+
+- I use an SSD and want to minimize disk writes
+- not flexible-enough to support my main goal - easy experimentation with
+ various ad-hoc monitors:
+ - I want to set different update intervals for different data sources
+ - I don't want long-running data collectors to block the main loop
+
+### Actuator
+Actuator is anything that takes action upon controller messages. A few generic
+ones are included:
+
+- `khatus_actuate_alert_to_notify_send`
+- `khatus_actuate_status_bar_to_xsetroot_name`
+
+and, by default, are left disconnected from the data feed, so if desired - need
+to be manually attached when starting `khatus`. See usage section.
+
+### Errors
+Any errors encountered by any sensor are propagated as alerts by the
+controller, which are in turn actualized as desktop notifications by the
+`khatus_actuate_alert_to_notify_send` actuator:
+
+![screenshot-self-error-propagation](screenshot-self-error-propagation.jpg)
+
+TODO
+----
+
+- status bar templating language
+- retry/cache for sensors fetching flaky remote resources (such as weather)
+- throttling of broken sensors (constantly returns errors)
+- alert specification language
+ - trigger threshold
+ - above/bellow/equal to threshold value
+ - priority
+ - snooze time (if already alerted, when to re-alert?)
+ - text: subject/body
+
+Redesign notes
+--------------
+
+- controller should not do formatting
+- need in-memory db for diskless feedback/throttling and cache
+- decouple sensor execution from sleep, i.e. a sensor is blocked not by sleep
+ process directly, but by reading of a pipe, to where a sleep process will
+ write a message announcing interval completion and thus signaling execution.
+ This will allow us to manually signal a sensor to update (concretely - I just
+ openned my laptop from sleep and want to force the weather to update
+ immediately); likewise, the sleep process should be blocked on pipe-read
+ until sensor execution is complete - this will allow us to reconfigure
+ intervals at runtime (which seems like a better idea than the above in-memory
+ DB one).