Moar ideas
[khatus.git] / README.md
index 7ab320b..cc4976d 100644 (file)
--- a/README.md
+++ b/README.md
@@ -2,78 +2,322 @@ khatus
 ======
 ![mascot](mascot.jpg)
 
-Experimental, system monitor and status (bar) reporter I use with
+Experimental system-monitor and status (bar) reporter I use with
 [dwm](https://dwm.suckless.org/) on GNU/Linux.
 
 ![screenshot](screenshot.jpg)
 
+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
-                |             |                 |
-                V             V                 V
- serial     +-----------------------------------------+
- stateful   |              controller                 |
- observer   +-----------------------------------------+
-                              |
-                          decisions
-                              |
-                              V
- serial     +-----------------------------------------+
- stateless  |              actuator                   |
- executor   +-----------------------------------------+
-                              |
-                        system commands
-                              |
-                              V
-                            ~~~~~~
-                            ~ OS ~
-                            ~~~~~~
+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 |
+              +----------------+              +-------------+
 ```
 
-### Actuator
-By default, actuator is left disconnected from the controller's output, so if
-desired - it needs to be manually attached when starting `khatus`. For example,
-in my `.xinitrc` I have:
+### 1.0
 
-```sh
-$BIN/khatus \
-2> >($BIN/twrap.sh >> $HOME/var/log/khatus.log) \
-| $BIN/khatus_actuator \
-    -v pid="$$" \
-    -v display=":0" \
-2> >($BIN/twrap.sh >> $HOME/var/log/khatus-actuator.log) \
-&
+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 ~~~~~~~~~~~~~~~~~
+            ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 ```
 
-(`twrap.sh` is a simple script which prefixes a timestamp to each line)
+### 0.x
 
-The idea is to later have multiple, (some more-general and some more-specific)
-actuators which can be selected as needed, say for example:
+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).
 
-```sh
-$BIN/khatus \
-| tee \
->(awk '/^STATUS_BAR/ {sub("^" $1 " *", ""); system("xsetroot -name \" " $0 " \"")}') \
->(grep '^REPORT' | actuate_report_to_email) \
->(grep '^ALERT' | grep mpd | actuate_alert_to_email) \
->(grep '^ALERT' | grep IntrusionAttempt | actuate_intruder_to_iptables_drop) \
->(grep '^ALERT' | grep NewDevice | actuate_alert_to_notify_send)
->(grep '^ALERT' | grep DiskError | actuate_call_mom)
-```
+This actually worked surprisingly-OK, but had limitations:
 
-... and/or any other such fun you might imagine.
+- 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
-actuator:
+`khatus_actuate_alert_to_notify_send` actuator:
 
 ![screenshot-self-error-propagation](screenshot-self-error-propagation.jpg)
+
+TODO
+----
+
+- tests (design is starting to take shape, so it is time)
+- show how many Debian package updates are available
+- show how many Debian package security-updates are available
+- monitor disk usage rate of change and alert if suspiciously fast
+- bring back CPU usage monitor
+- actual METAR parser, to replace the flaky `metar` program
+- 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
+- monitor processes
+    - totals (grand and per state)
+    - zombies
+    - threads
+    - CPU hogs
+    - memory hogs
+    - memory leaks (if some process consistently grows)
+    - is select process up?
+    - log resource usage of select processes
+- monitor arbitrary HTTP endpoint availability
+    - is status within expected range?
+    - response time
+        - is responce time within acceptable range?
+- report detailed status upon request (to a terminal)
+    - use color to indicate age of data
+- monitor logins
+    - totals (per time period)
+        - failures
+        - successes
+    - most recent
+        - success
+        - failure
+- monitor battery time remaining
+    - monitor accuracy (is percentage change rate on track to meet estimate?)
+    - adjust estimate based on observed inaccuracies in past estimates (Kalman?)
+
+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).
+
+Ideas
+-----
+
+- daemonize `khatus`, so we don't have to re-launch `X11` to re-launch `khatus`
+- interoperate with other khatus instances
+    - prefix machine ID to each data source
+      (What should that ID be? Hostname? Pub key?)
+    - fetch remote data and process locally
+    - what transport to use?
+        - ssh + rsync + cache dumps per some interval?
+    - `A` can setup self penetration testing, by setting up probe of `A` on `B`
+      and fetching results from `B` to `A`
+- offline mode - quick disable all network-using subsystems (sensors, monitors, etc)
+- classify each sensor as either "local" or "remote" (what about `iwconfig`, et al?)
+- store data with rrdtool
+- some kind of personal calendar thing integration
+- monitor tracking numbers (17track should be easiest to get started with)
+- monitor password digests against known leaked password databases
+- monitor stock prices
+- monitor some item price(s) at some store(s) (Amazon, etc.)
+    - https://docs.aws.amazon.com/AWSECommerceService/latest/DG/EX_RetrievingPriceInformation.html
+    - https://docs.aws.amazon.com/AWSECommerceService/latest/DG/ReturningPrices.html
+    - https://developer.amazonservices.com/
+- monitor Amazon order status
+    - https://developer.amazonservices.com/gp/mws/api.html?group=orders&section=orders
+- monitor eBay order status
+    - http://developer.ebay.com/DevZone/XML/docs/Reference/eBay/GetOrders.html
+- monitor eBay auctions (https://en.wikipedia.org/wiki/EBay_API)
+- monitor PayPal (https://www.programmableweb.com/api/paypal)
+- monitor bank account balance and transactions
+    - https://communities.usaa.com/t5/Banking/Banking-via-API-Root/m-p/180789/highlight/true#M50758
+    - https://plaid.com/
+    - https://plaid.com/docs/api/
+    - https://plaid.com/docs/api/#institution-overview
+    - https://github.com/plaid
+    - https://www.bignerdranch.com/blog/online-banking-apis/
+- monitor/log road/traffic conditions
+    - travel times for some route over a course of time
+    - https://msdn.microsoft.com/en-us/library/hh441725
+    - https://cloud.google.com/maps-platform/
+    - https://cloud.google.com/maps-platform/routes/
+    - https://developer.mapquest.com/documentation/traffic-api/
+    - https://developer.here.com/api-explorer/rest/traffic/traffic-flow-bounding-box
+- monitor news sources for patterns/substrings
+    - http://developer.nytimes.com/
+    - https://news.ycombinator.com/
+    - https://lobste.rs/
+    - https://www.undeadly.org/
+    - http://openbsdnow.org/
+    - https://lwn.net/
+- monitor a git repository
+    - General
+        - total branches
+        - age of last change per branch
+        - change set sizes
+    - GitHub
+        - pull requests
+        - issues
+- monitor CI
+    - Travis
+    - Jenkins
+- pull/push data from/to other monitoring systems (Nagios, Graphite, etc.)
+- monitor file/directory age (can be used for email and other messaging systems)
+- monitor mailboxes for particular patterns/substrings
+- monitor IRC server(s)/channel(s) for particular patterns/substrings (use `ii`)
+- monitor iptables log
+    - auto-(un)block upon some threshold of violations
+- monitor changes in an arbitrary web resource
+    - deletions
+    - insertions
+    - delta = insertions - deletions
+- monitor/log LAN/WAN configurations (address, router, subnet)
+- monitor/log geolocation based on WAN IP address
+- correlate iptables violations with network/geolocation
+- monitor vulnerability databases
+    - https://nvd.nist.gov/
+    - https://vuldb.com/
+    - http://cve.mitre.org/
+- vacation planning optimization
+    - I want to visit a set of places within some time period. Given the
+      current set of prices, a set of constraints (I need to stay some amount
+      of days at each, I must be in X at Y date, etc), which visiting dates for
+      each are cheapest?
+- browse https://www.programmableweb.com/ for some more ideas
+- GC trick: instead of actually doing GC, do a dummy run of building a status
+  bar at `BEGIN`, to fill-in the atimes for keys we need, then use the atimes
+  keys to build a regular expression to accept messages only from keys we
+  actually use
+
+Many of the above will undoubtedly need non-standard-system dependencies
+(languages, libraries, etc.), in which case - would they be better off as
+separate projects/repos?
+
+With all these ideas, it is starting to sound very noisy, but no worries - to
+quickly and temporarily shut everything up - just kill `dunst` and or toggle
+the status bar (`Alt` + `B` in `dwm`). For a permanent change - just don't
+turn-on the unwanted monitors/sensors.
This page took 0.029304 seconds and 4 git commands to generate.