dht | Distributed hash table using a chord

The documentation for mariadb's datetime states that 0 microseconds is the default datetime precision. This appears to be so in your db output, ie. 2022-03-26 10:00:11. Also it appears from the same documentation that the time zone depends on the session.

I was able to recreate this exception as well as a half-fix by using DATETIME(fsp=6). This is what appears to happen:

1. DHT is created using python's datetime.now() as the primary key in your application
2. DHT is written to the db, the microseconds are probably just ignored by mariadb because the column is expecting no microseconds by default
3. refresh is called, SQLAlchemy queries the db looking for a record with datetime = 2022-03-26T17:00:11.815 (note there are microseconds here)
4. it can't find one because the datetime in the db trimmed those down to 2022-03-26T17:00:11, so refresh fails with an exception

Right here I'd say the safest strategy would be to use an integer id for the primary key but ...

To make this work first you could bump the precision up on the column to 6 places for microseconds. You have to use the dialect specific DATETIME, from sqlalchemy.dialect.mysql import DATETIME, with fsp=6.

Then timezones... you should make sure the db session time zone matches the time zone you use to make the datetime in your app because mariadb does not do conversion. UTC is probably the safest thing here but I'm not sure of the specifics of mariadb.

You might have to change the server's default itself. This might work for the session, session.execute(text("SET time_zone = 'UTC'")).

In python you could do default = lambda: datetime.datetime.now(datetime.timezone.utc).replace(tzinfo=None) via import datetime. This will get the current time in UTC but then remove the timezone to make the dt naive.

A template processor is used in:

You can use ipdb for debugging in the console.

So finally after hours of debug I found the reason and solution.

First of all, I debugged size of the set each time I open it. And weirdly after first call its size becomes 0, so set is erased after the first call to ignite.set(). After this I switched to plain cache (instead of set) and just check cache.containsKey(user). Its size was persistent among the getOrCreateCache() calls but the NPO problem was still raised.

Then I found this tiny little answer on Ignite mailing list where it's said that Ignite caches implement AutoCloseable interface. Which means that after try-except block cache.close() is automatically called. And it means not just close the "connection" to cache but stopping the cache itself.

After this I changed my code to this:

The problem here is not that you are doing computations on a requires_grad=True tensor. After all this is how one gets gradients! By doing computations on such tensors :)

The issue is that you are doing what are called in-place operations.

By in-place we mean that a previously existing variable's memory location is now replaced with a new variable. As a result the computational graph is broken, and thus, no gradient backpropagation can be achieved.

How does this look like? I found a few quick examples in this Pytorch-forum question

In particular:

The first thing you should notice is that sending numbers throug the serial interface will result in different string lenghts depending on the number of digits. So reading a fixed number of 6 bytes is not a good idea. (actually this is almost never a good idea)

You terminate each sensor reading with a linebreak. So why not use readline instead of read[6].

Here v1 = int(float(i[1:5].strip('\\r\\nr'))) you're trying to remove \r, \n and r from the received string. Unfortunately you escaped the backslash so you're actually stripping \, r and n. \r is actually something where you need the backslash to represent the carriage return character. Don't escape it!

In the first run loop() will send something like:

r 0.00\r\nh 40\r\nc 25\r\n

So the first 6 bytes are r 0.00. So i[1:5] is 0.0. As you see there is nothing to escape. Also 5 is excluded so you would have to use i[2:6] to get 0.00. But as mentioned above using fixed lenghts for numbers is a bad idea. You can receive anything between 0.00 and 100.00 here.

So using readline you'll receive r 0.00\r\n

The first and last two characters are always there and we can use [2,-2] to get the number inbetween regardless of its length.

Something like this?

Replace

Your code defines a username and password for the MQTT broker but never actually uses them. You need to pass them to the connect method or else they do nothing.

Okay so thanks to the input of Mat I tried substituting the DHT11 library with something more sleek, which took me a while to get up and running. I ended up using this as a base, edited around a bit and commented heavily for my benefit. I added my updated code below for anyone interested (thanks for pointing out the correct highlighting issue), there's also a github with the rest of the design files.

Seems the library is really heavy, as the compiler output shows:

Compiler output: