Asta Areti

@a43ti.bsky.social

33 Followers 26 Following

Programming recent graduate, nearly every day surprised by something new from professional realm .NET | GIS | DevOps | Angular/React | Backend | Android github 👉👉👉 https://github.com/astk4 tg: https://t.me/astareti tt https://www.tiktok.com/@asta.a43t1

current frontend dockerfile, in which, as you can see, the contents of former "container setup" script are here directly in a RUN. connected with && to keep it 1 layer despite several commands. and so I can launch the actual program (nginx) in a more classic way - dedicated CMD instruction. that new script "fetch_frontend_script" before npm install - it downloads and unzips libraries which I couldn't install with npm, so had to put into /assets/ folder

initial problematic situation: during and after project usage, despite backend working completely successfully, especially the microservice that listens to AIS messages stream, sends the significant ones to main API, and what's important here - stores them in history database table. but the latter - as you can see, it didn't

error messages at every attempt of sending data to main backend via grpc with configuration how it was before (only one port). apparently it happened in docker because in the network between containers I'm using http. not https, therefore no protocol negotiation that can change http version 1.1 to 2. unless explicitly specified http2 (probably it was mixed mode), in such situation it is assumed that all http requests are of version 1.1. but grpc requests are by definition over http2, so they are conflicting with this port

docker-compose fragment describing postgre sql container. especially look at the volumes part

frontend dockerfile. basically it pulls my angular app into a container with Node 20, and builds it there to get static html + css + js. these are hosted on Nginx web server which is another layer of this container, that's why angular/npm build output is copied into nginx folder. and my custom steps are to copy my additional custom config for Angular (spoiler - regarding integration with backend) into container, and whatsv even more interesting - copy bash script (which contains even more commands to move files around, see next pic), clean it from windows-specific text artifacts (yes thats the scary line 17) and run it

demo comparison of behaviour of the settings page in situations that invoke activity recreation. these random colors you see in the "before" part are from material you - a feature in modern android versions that chooses accent color for UI as ± average color of your wallpaper

$comparison of old and new sizing setting mechanics ¯⁠\⁠_⁠(⁠ツ⁠)⁠_⁠/⁠¯$

updated workflow! as you can see, after dealing with sdk and gradle there is getting keystore (from a base64 github secret), then within the stage of building release APK a key is gotten from the keystore, and with the key APK is signed

the new workflow! added building the cpp lib, uploading it as its own artifact, and copying it to wpf app directory! :) next steps as before - wpf in a release configuration, upload the resulting archive as artifact too

tasks of solution's development in jira (not in the kanban board form because almost everything I wanted is done and on a board instead of a table it would look inconvenient)

bluetooth configuration mobile activity before the fix, where in both lists there are not only potential laptops/computers but also tvs, earphones and speakers...

workflow results popup on a commit, that appears when we click near the commit message on the indicator of a success (like here) or lack thereof, which appears when there are any workflow yamls in the repo

workflow results popup on a commit, that appears when we click near the commit message on the indicator of a success (like here) or lack thereof, which appears when there are any workflow yamls in the repo

"before" section is an old version, see how charging triggers a clean but not asked for disconnection. "after" section is the freshly fixed version. as you can see, the... mouse pretending session stays fully the same - connection, interface layout including screen dimensions, even the selected button. also didn't think I would have to make a layout for portrait orientation in the touchscreen activity but I had to, for the app to not look weird on some ROMs for a short moment of... that forced vertical state when system charging animation is played

not connected to a phone yet, trying to discover visible bluetooth devices

bluetooth configuration page. yes, thats on the other phone, API 31, than the API 25 one in the video, hence the many visible bluetooth devices while in the video there were discovered none. but showing y'all this just to show how UI looks like for the visible ones. namely, the arrow button click pairs the device with the phone

demo how to use the apps. the desktop part is launched first. select a bluetooth device and start waiting for connection from it, which is begun by the mobile app. in the mobile app you should choose the pc/laptop/whatever desktop device that's waiting. and go to the touchscreen where the main functionality happens. here I am demonstrating "mouse" move, clicks, double clicks, long press and dragging that allows to draw,with the ability to switch left/right button, which are done by taps/double taps/long press and drag on screen; and "wheel" scrolling which is done with volume buttons: volume + is up, - is down. after finishing ur mouse simulation you should disconnect, which can be done both from phone as shown and from the desktop part. and that nice little close confirmation window is added so that one doesnt close the program with an accidental mis-"click" on the phone in the wrong (close button) place.

disk C space usage chart before shrinking the docker virtual hard disk (vdhx) file, which had an obscene size of 17 GB

2 queues necessary for the data exchange during the batch image processing, on a cloud rabbitmq instance. btw, cloudamqp.com I recommend y'all. so, thumbnails_tasks queue is for the messages with original path and max thumbnail dimension value, aka orders for scaled images. the thumbnails_results queue is for the messages with only paths of the resulting thumbnails, which are sent to confirm that the scaling is completed and the main program can chill out :))

demo video of mass creating thumbnails for a batch of images. first here the several instances of programs that actually scale images are launched. delays are purely for demonstration purposes. then launched is the program that sends to rabbitmq (in this case in cloud) messages about which images should be made thumbnails for. and the scaling program instances read these messages from rabbitmq queue. finally, thumbnail examples are shown, original images are preserved as you can see

The general architecture is a hybrid between client-server and microservice. main server part is an ASP. NET Core Web API, which contains most of AIS to visualization pipeline, and regular REST endpoints for getting other data - EEZ and ports. a microservice, which is a console app, communicates via gRPC with it. processing AIS messages involves both reading and writing to Redis. functionality of the REST endpoints is getting (there's only getting! not whole CRUD!) the port points and EEZ GeoJSON stored in PostgreSQL with PostGIS extension. Also the microservice is connected, figuratively speaking - on the other end, to a websocket from where AIS is streamed. This app constantly receives new messages, parses them, filters out only types relevant for visualization and sends them to main backend asap. Protobuf types for GRPC communication and the classes generated from them are contained in a separate class library project.

proper way to implement idisposable interface in the full way, proposed by IDE (from maritime project)

see, no repetitions in the first 2 columns - same amount of distinct session IDs as that of start times. if something was off in saving logic, with DISTINCT command there would be weird records like several "starts" for the same session ID, or vice versa

confirming the amount of rows, and data size as well, I wasn't lied to about how good clickhouse is at data compression 😎

initial exception I suddenly started getting yesterday out of the blue, without recent code changes here, while trying to connect to AIS websocket as usual with default options

this is done by tools -> detect scene changes -> input the coefficient, the smaller its value the more sensitivity to frame changes, therefore the more scenes

consumer thread in progress where batches of AIS messages are gathered. as u can see, the main interval wait (the 10 seconds) is done with periodic timer, the rest if necessary - with empty loop. the values of min interval duration and min batch size are specified in appsettings. the countercontainer array btw, if ya'll are curious, is a workaround to imitate passing a ref int parameter to async method

first half of the table creation query. table will have columns for message ordering in time, and then properties of different AIS message types are all implemented here in one table too. I know that normally in such situations table JOIN is used, but with near-real-time replay constant JOINing would seriously harm the performance. so I'm making the table like already a result of a join. cells in a row that do not belong to the current message type, will have null or 0

clickhouse image in question. weird, I thought, working with it will definitely involve ports... hmmmm

what can i say, if i saw this earlier it would save me a couple days of putting the streaming service code upside down

the "consumer" part of the attempted architecture, as you can see, HashGetAll (HGETALL) is called for every single object...

how the data organization looked like in redis: lots of vessel hash tables, basically AIS grouped by MMSI, floating directly in DB

elapsed time (measured with stopwatch) of processing AIS ship data streams... in the attempted architecture in question - obviously it surpasses the periodic timer to an inadmissible extent

connection error in question, this printed out instead of every benchmark

in the modified circle equation, the key to separation is the subtraction under the sqrt

and cos function is modified to be mirrored in both dimensions and align with the semicircles, nothing scary

previous problematic situation, markers closer to the poles were increased in size, just like anything else in mercator projection

demo of the marker sizing in different places on the map - from triangle nicely matching with the dots, to l o n g needles

encountered problem with java compiler settings. at this point I already set JDK 21 for the project, module and in maven pom.xml - and successfully coded using the new apis. but during build attempt there were errors exclusively due to the new methods from jdk21 not being recognized (last pic in this post). then I learnt that in this case one should check language level in compiler settings and set it to the respective value (21 in my case). but it wasn't possible here((0(

the moment in the logs of one of CVAT containers, where there was the same connection error repeating, already something not about wrong credentials, untill I created the cvat database myself manually

main changes in the main docker-compose file of CVAT: deleted configurations of postgres (cvat_db) and redis containers; environment variables for connection to postgres went to the main anchor (where variables are gathered to avoid repeating) and on this level CVAT_ was added to them. apparently, as I made the conclusion from logs - to override the default connection config, which was adjusted for server and DB created by the postgres instance from the container, that I deleted

prior content of the hash table with ships' coordinates. these were objects with latitude and longitude properties, encoded into bytes (with messagepack...) and stored under MMSI (=real world id) of a respective ship. hash table with type ids was more readable, because those are simple values, but still, it was separately...

initial problematic situation - I joined an edge loop into a face, and got this weird warp, initially I suspected that the problem was with normals

video demo of how requests in this project travel from MVC actions - to http handler - then to backend REST API - sometimes to http handler again - back to MVC actions returning views (so many interesting moments that I didn't try to convey it with 4 images and just recorded)

service-level method where a refresh token is replaced. in other implementations, when refresh tokens have arbitrary IDs, this subtask can be done with an update operation. but as mine correspond to JWT token IDs, and refresh is by definition called when a jwt is outdated, as well as we can't update key properties with ef core - i gotta delete an entity with the old jwt, and add a new entity where everything is new except for the corresponding user ID

a successful /refresh request inspected in browser devtools: input request has a refresh token cookie, result response sets this cookie to a new value.

example of some tables breakdown in pg4admin gui, demonstrating how there are NO indexes created on primary keys shown 0_o (in the first table see, the "indexes" tab is opened too)

service method that creates a jwt token. no jwt/identity specific types in input or output (thats my viewmodel). claims in the token include not only obvious user properties but also Sub(ject), user id - I didn't use it before - and Jti, token id. token's time properties, validity start and expiry, are saved in utc (earlier inexperienced me used local time, dont do that)

the text file with log records of different types from the api's work. the numbers in square brackets are EventIds btw

demo video of the maritime project focusing on ships positions markers, which are now colored, in different places over the world