Noticed that in this video where we are enqueue'ing 1e6 number and dequeuing those 1e6 inserted numbers won't actually result in queue size being 0 at the end Because we are doing wgE and wgD together, if the dequeue operations happen more rapidly than enqueue operations then that can lead to queue size at the end being > 0
One more important(and necessary) feature to add in thread safe queue is waiting and notification using condition variables! So that consumer threads will wait until there is something to read! And producer threads will wake up consumers when they enqueue something
It’s a nice intro but there are caveats. For example, batch processing which you mentioned in the example, it’s better to use multiple processes and leverage parallelism than managing a complex thread safe code. Considering compute is quite cheap, parallelism >>> concurrency unless exceptions And also there are alternatives to muteness where we leverage hardware locks using atomic variables in c++ and that’s more preferred compared to mutex. Sure, it’s not a magic bullet but definitely it has its own pros
Seeing how you connect your previous videos to current ones is great. Please make a video on TCP Queue. It would make an awesome connecting video. I have done the implementation on Rust, eager to see how you implement in Go and what is your thought process when designing it.
This was nice, I use elixir which is a functional language so their is no mutable shared state and therefore there are no problems related to thread safety. Every process gets its own copy of memory and works in isolation, processes exachange information via message passing. Its a different approach to solve thread based problems without using locks. Companies like whatsapp and discord are using it at sacle so it actaully works.
Hi Arpit, Doesn't applying mutex on threading makes the program equivalent to a program running sequentially? Why should I use threading and mutex instead of simple sequential code in such scenarios?
I think it is more like the function will be added to the stack after the before function stops execution in the case of sequential code though in this case the function has already been loaded on to the stack and is just waiting for execution. This is my analogy, not sure though :)
I have an doubt/Question .... Arpit - when we do use multithreading and mutexs, if I think on a high level it makes an "so called asynchronous program" to be act as a synchronised program?. When if it threads/routines with mutexes eventually making the execution synchronous on multiple workers how does it improving the performance and speeding up the execution in a multi processor environment??. May be i messed up so many questions into one single comment. would be helpful it you can help me to understand the background processes?.
This is a very nice question and you are actually correct. Having locks essentially makes access to the data structure synchronous. The reason why locks are helpful is to avoid inconsistencies when multiple threads try to update the same memory location as shown in the video. When you use locks you might end up losing the performance benefits of concurrent programming since threads might be waiting on a lock
Usually we only use locks in critical sections which can be a small set of instructions, imagine a scenario where we are fetching something from db thendoing some work on a shared variable. We'll only require the lock on this part where we are accessing the shared variable so the db calls can happen parallely thereby improving performance.
@@shervilgupta92 I think you nailed the answer when you mentioned 'critical section'. Coming back to questions: It takes some doing to design and code in a way that critical section is, first of all, identified & isolated and rest of things are structured around it. This 'critical section' needs safety. One way (easy and no-brainer) is to use 'pessimistic locking' as shown in video. In real world, it would cover most of the scenarios that one would work on, anything smart would definitely need a lot more scrutiny. You don't want call at middle of night due to that smartness. If your critical section is too big i.e. whole method then, yeah, its performance issue. You end up serializing whole thing and there won't be performance benefit. It may help to try things with thread safe data structures, immutability and so on. But there is no easy way out unless some restructuring is done on design and approach itself.
Got the issue . Since I was trying to get the size of the queue just after the for loop of enqueue : there were still some elements to be inserted in the queue (These go routinues were still not scheduled on actual thread by go scheduler ) . Due to this we were getting incorrect queue size . `var i int32 = 0 for i < NUM_THREADS { atomic.AddInt32(&i, 1) wgE.Add(1) go func() { queue.Enqueue(rand.Int31()) wgE.Done() }() } fmt.Println("just after enqueue size:", queue.Size(), "....", i) fmt.Println("size:", queue.Size(), "....", i)`
Hi everyone if anyone can help me i have written a not a thread safe code but still its acting like thread safe can anyone help me here i tried to find the reason but not able to do so, if any one can explain here the code in python import threading import random class ConcurrentQueue: def __init__(self): self.queue = [] def enqueue(self, item): self.queue.append(item) def size(self): return len(self.queue) NUM_THREADS = 1000000 def main(): queue = ConcurrentQueue() def enqueue_task(): queue.enqueue(random.randint(1, 1000000)) # Create and start enqueue threads enqueue_threads = [threading.Thread(target=enqueue_task) for _ in range(NUM_THREADS)] for thread in enqueue_threads: thread.start() # Wait for enqueue threads to finish for thread in enqueue_threads: thread.join() print("size:", queue.size()) if __name__ == "__main__": main()
Python is single threaded. It is interleaving the execution and not really leveraging multiple cores to lead to sync issues in values. Would recommend switch it C, Java, or Go.
Please reply i am asking this from last 2-3 days Hi arpit , I work at startup didn't joined any college, i am self taught programmer , now also want to learn stuffs like complier, operating systems, can i learn it from resources available online or do i need to join college for this. 😮
@AsliEngineering thank you so much One request please bring content on computer architecture I mean internals topics of computer science, everyone is bringing content like teaching frameworks, libraries these are very high level, you please do something different like advanced part of computer science Topics like Design of compilers, Design of operating system Everyone lacks in these parts
Noticed that in this video where we are enqueue'ing 1e6 number and dequeuing those 1e6 inserted numbers won't actually result in queue size being 0 at the end
Because we are doing wgE and wgD together, if the dequeue operations happen more rapidly than enqueue operations then that can lead to queue size at the end being > 0
One more important(and necessary) feature to add in thread safe queue is waiting and notification using condition variables!
So that consumer threads will wait until there is something to read! And producer threads will wake up consumers when they enqueue something
I wish I had discovered this channel during my engineering.
Same here
same here..
It’s a nice intro but there are caveats. For example, batch processing which you mentioned in the example, it’s better to use multiple processes and leverage parallelism than managing a complex thread safe code. Considering compute is quite cheap, parallelism >>> concurrency unless exceptions
And also there are alternatives to muteness where we leverage hardware locks using atomic variables in c++ and that’s more preferred compared to mutex. Sure, it’s not a magic bullet but definitely it has its own pros
Seeing how you connect your previous videos to current ones is great. Please make a video on TCP Queue. It would make an awesome connecting video. I have done the implementation on Rust, eager to see how you implement in Go and what is your thought process when designing it.
Thanks for suggesting. Added it to my list to topics to cover.
This was nice, I use elixir which is a functional language so their is no mutable shared state and therefore there are no problems related to thread safety. Every process gets its own copy of memory and works in isolation, processes exachange information via message passing. Its a different approach to solve thread based problems without using locks. Companies like whatsapp and discord are using it at sacle so it actaully works.
That's why memory ussage goes 🚀
It actually uses some tricks like persistent data structures to avoid copying memory.
Your content is a gold mine.
Hi Arpit, it would be great if you can explore the topic of lock free data structures
Hi Arpit,
Doesn't applying mutex on threading makes the program equivalent to a program running sequentially? Why should I use threading and mutex instead of simple sequential code in such scenarios?
Same doubt! Did you find answer?
I think it is more like the function will be added to the stack after the before function stops execution in the case of sequential code though in this case the function has already been loaded on to the stack and is just waiting for execution. This is my analogy, not sure though :)
Please add a connecting video on how queue used inside kafka
I reallly love your videos! keep it up
I have an doubt/Question ....
Arpit - when we do use multithreading and mutexs, if I think on a high level it makes an "so called asynchronous program" to be act as a synchronised program?.
When if it threads/routines with mutexes eventually making the execution synchronous on multiple workers how does it improving the performance and speeding up the execution in a multi processor environment??.
May be i messed up so many questions into one single comment. would be helpful it you can help me to understand the background processes?.
I have the same doubt.
This is a very nice question and you are actually correct.
Having locks essentially makes access to the data structure synchronous. The reason why locks are helpful is to avoid inconsistencies when multiple threads try to update the same memory location as shown in the video.
When you use locks you might end up losing the performance benefits of concurrent programming since threads might be waiting on a lock
Usually we only use locks in critical sections which can be a small set of instructions, imagine a scenario where we are fetching something from db thendoing some work on a shared variable. We'll only require the lock on this part where we are accessing the shared variable so the db calls can happen parallely thereby improving performance.
When your program is IO bound, use concurrency. If it’s not, use parallelism
@@shervilgupta92 I think you nailed the answer when you mentioned 'critical section'.
Coming back to questions:
It takes some doing to design and code in a way that critical section is, first of all, identified & isolated and rest of things are structured around it. This 'critical section' needs safety. One way (easy and no-brainer) is to use 'pessimistic locking' as shown in video. In real world, it would cover most of the scenarios that one would work on, anything smart would definitely need a lot more scrutiny. You don't want call at middle of night due to that smartness.
If your critical section is too big i.e. whole method then, yeah, its performance issue. You end up serializing whole thing and there won't be performance benefit. It may help to try things with thread safe data structures, immutability and so on. But there is no easy way out unless some restructuring is done on design and approach itself.
Would love to see a performance comparison between atomic CAS based implementation and this lock based one
Thanks for suggesting. Video on it dropping soon.
Very informative 👏 😊
Tried the same code and after enqueue , getting size of queue as 999987 instead of 1000000. Even after using Locks . Can someone please help me here ?
Got the issue . Since I was trying to get the size of the queue just after the for loop of enqueue : there were still some elements to be inserted in the queue (These go routinues were still not scheduled on actual thread by go scheduler ) . Due to this we were getting incorrect queue size .
`var i int32 = 0
for i < NUM_THREADS {
atomic.AddInt32(&i, 1)
wgE.Add(1)
go func() {
queue.Enqueue(rand.Int31())
wgE.Done()
}()
}
fmt.Println("just after enqueue size:", queue.Size(), "....", i)
fmt.Println("size:", queue.Size(), "....", i)`
wow, insightful. Thankyou for the video
Keyboard sound is very irritating so please remove them in features video
Hi everyone if anyone can help me
i have written a not a thread safe code but still its acting like thread safe can anyone help me here
i tried to find the reason but not able to do so, if any one can explain
here the code in python
import threading
import random
class ConcurrentQueue:
def __init__(self):
self.queue = []
def enqueue(self, item):
self.queue.append(item)
def size(self):
return len(self.queue)
NUM_THREADS = 1000000
def main():
queue = ConcurrentQueue()
def enqueue_task():
queue.enqueue(random.randint(1, 1000000))
# Create and start enqueue threads
enqueue_threads = [threading.Thread(target=enqueue_task) for _ in range(NUM_THREADS)]
for thread in enqueue_threads:
thread.start()
# Wait for enqueue threads to finish
for thread in enqueue_threads:
thread.join()
print("size:", queue.size())
if __name__ == "__main__":
main()
Python is single threaded. It is interleaving the execution and not really leveraging multiple cores to lead to sync issues in values.
Would recommend switch it C, Java, or Go.
@@AsliEngineering thank you
👍🏼👍🏼
Please reply i am asking this from last 2-3 days
Hi arpit ,
I work at startup didn't joined any college, i am self taught programmer , now also want to learn stuffs like complier, operating systems, can i learn it from resources available online or do i need to join college for this.
😮
you can learn everything online. just put in the required efforts.
@AsliEngineering thank you so much
One request please bring content on computer architecture I mean internals topics of computer science, everyone is bringing content like teaching frameworks, libraries these are very high level, you please do something different like advanced part of computer science
Topics like
Design of compilers,
Design of operating system
Everyone lacks in these parts
@@pritamkumar6454 bruh go to collage
Ahh
Devs training llm to outdate themselves of future jobs ✌️✌️
Classic