This repository was archived by the owner on Apr 14, 2022. It is now read-only.
Repository navigation
This repository was archived by the owner on Apr 14, 2022. It is now read-only.
Python Language server takes a huge amount of RAM (more than 10+GB) #1426
Description
Today I faced this issue! it took more than 10GB+ of my system RAM. I cant reproduce it now (see below), it seems when this happens, the intellisense also stops working. and analysis goes on forever!
I seemed to me, this happens when I try to code (use intellisense) when the language server is still analyzing something, and that just kills the intellisense for good! and each time I try to write new code, and analysis continues, more ram is consumed. I'm not sure about this but I got this feeling.
I attached some log files, hopefully they should show something.
I'm using the latest version
Python Language Server:
Microsoft Python Language Server version 0.3.46.0
Python Extension :
2019.8.29288 (6 August 2019)
VSCode info
Version: 1.37.0 (user setup)
Commit: 036a6b1d3ac84e5ca96a17a44e63a87971f8fcc8
Date: 2019-08-08T02:33:50.993Z
Electron: 4.2.7
Chrome: 69.0.3497.128
Node.js: 10.11.0
V8: 6.9.427.31-electron.0
OS: Windows_NT x64 10.0.17763
UPDATE:
OK this happened again and I could save the log and record it before the vscode crashed!
Here is the console log :
Here is the second vslog and the the language server log (copied form output tab):
download 2 log files separately .
and this is the sample test code I used :
#%%
# in the name of God
import torch
import torch.nn as nn
import torchvision
#import torchvision.transforms as transforms
from torchvision import transforms
from torchvision import datasets
from torch import cuda
# Device configuration
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(f'training happens on : {device}')
# Hyper parameters
num_epochs = 5
num_classes = 2
batch_size = 100
learning_rate = 0.001
# number of threads for faster IO
num_workers = 2
# MY dataset
transformations = transforms.Compose([transforms.Resize(256),
transforms.ToTensor(),
transforms.Normalize(mean=(0.5, 0.5, 0.5),
std=(0.5, 0.5, 0.5))
])
train_dataset = datasets.ImageFolder(root=r'C:\Users\Mariane\Desktop\Testpy\train_data',
transform=transformations)
# you didnt provide a test data so we are using the training data for the sake of
# our experiment!
test_dataset = datasets.ImageFolder(root=r'C:\Users\Mariane\Desktop\Testpy\train_data',
transform=transformations)
# Data loader
train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
batch_size=batch_size,
shuffle=True,
num_workers=num_workers)
test_loader = torch.utils.data.DataLoader(dataset=test_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=num_workers)
# Convolutional neural network (two convolutional layers)
def conv_bn_relu(in_channel, out_channel, kernel_size=3, stride=1, padding=1,
bias=False, batchnorm=True):
layers = nn.ModuleList()
conv_layer = nn.Conv2d(in_channels=in_channel, out_channels=out_channel,
kernel_size=kernel_size, stride=stride, padding=padding, bias=bias)
layers.append(conv_layer)
if batchnorm:
layers.append(nn.BatchNorm2d(out_channel))
layers.append(nn.ReLU())
return nn.Sequential(*layers)
# for using functional form of classes and methods
import torch.nn.functional as F
class ConvNet(nn.Module):
def __init__(self, num_classes=10):
super(ConvNet, self).__init__()
self.net = nn.Sequential(conv_bn_relu(3, 16, 5, 1, 2),
nn.MaxPool2d(kernel_size=2, stride=2),
conv_bn_relu(16, 32, 5, 1, 2),
nn.MaxPool2d(kernel_size=2, stride=2)
)
# method 2
# self.layer1 = conv_bn_relu(3, 16, 5, 1, 2)
# self.layer2 = conv_bn_relu(16, 32, 5, 1, 2)
# self.maxpool = nn.MaxPool2d(2, 2)
self.fc = nn.Linear(32, num_classes)
def forward(self, x):
# first method
out = self.net(x)
# second method :
# out = self.layer1(x)
#print(f'layer1: output.shape : {out.shape}')
# out = self.maxpool(out)
#print(f'maxpool: output.shape : {out.shape}')
# out = self.layer2(out)
#print(f'layer2: output.shape : {out.shape}')
# out = self.maxpool(out)
#print(f'maxpool: output.shape : {out.shape}')
out = F.avg_pool2d(out, out.size()[2:])
out = out.view(x.size(0), -1)
# print(out.shape)
out = self.fc(out)
return out
model = ConvNet(num_classes).to(device)
# Loss and optimizer
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
# Train the model
total_step = len(train_loader)
for epoch in range(num_epochs):
for i, (images, labels) in enumerate(train_loader):
images = images.to(device)
labels = labels.to(device)
# Forward pass
outputs = model(images)
loss = criterion(outputs, labels)
# Backward and optimize
optimizer.zero_grad()
loss.backward()
optimizer.step()
if (i+1) % 100 == 0:
print('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}'
.format(epoch+1, num_epochs, i+1, total_step, loss.item()))
# Test the model
model.eval() # eval mode (batchnorm uses moving mean/variance instead of mini-batch mean/variance)
with torch.no_grad():
correct = 0
total = 0
for images, labels in test_loader:
images = images.to(device)
labels = labels.to(device)
outputs = model(images)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).float().sum().item()
print('Test Accuracy of the model on the 10000 test images: {} %'.format(
100 * correct / total))
# Save the model checkpoint
torch.save(model.state_dict(), 'model.ckpt')
```
Activity
jakebailey commentedon Aug 12, 2019
I can see a load of exceptions in those language server logs; I think it might be failing halfway through and restarting. Can you try out the beta build to see if the issue goes away? Set the following in your user settings (ignore the warning that the option doesn't exist), then reload.
I think the exceptions are #1335, which is fixed in 0.3.47+, which I have now moved to stable as well.
Coderx7 commentedon Aug 13, 2019
The intellisense doesnt work anymore after I enabled this. I didn't encounter a leak this time, (I dont know if this is related to enabling the beta or not (as I haven't faced one with the old version either after this), so far I have not faced any leaks
However, the analysis kept working and never finishes. what I noticed was that, whenever I pressed a key to type something, the analysis report would print :
so basically if I stopped pressing any key it would stop, and if I continued, it would do so as well(even writting in a python string as in
datasets.ImageFolder('ssssssss')would trigger this).The intellisense at this rate is long gone and there is no intellisense.
I removed the beta related setting and intellisense came back on (after I restarted ) (Also I'm not facing a leak now, leaking seems to happen at random at this rate)
Here are the logs for when the beta was active : logfile
jakebailey commentedon Aug 13, 2019
In the same environment, can you open a file that's just
import decorator, and see if you see an exception come up when doing something likedecorator.to trigger completion on the module?Coderx7 commentedon Aug 14, 2019
yes I do here is the log :
log
jakebailey commentedon Aug 14, 2019
Based on that, I'm fairly certain this is the same as #1401 (but we can reopen as needed).
I'd appreciate it if you could set the following in your config and reload to trigger a download (to v0.3.56), then see if you can grab one of the stack traces for those exceptions, and post an update on #1401.
Closing in favor of #1401.
Coderx7 commentedon Aug 26, 2019
it happened again just moments ago! it maxed out 10+G of system RAM. unfortunately I couldn't save any log!
Coderx7 commentedon Aug 31, 2019
The leak happened moments ago again! and here is the log : https://we.tl/t-7XyjjmUIgf
MikhailArkhipov commentedon Aug 31, 2019
Seyyed Hossein Hasanpour (@Coderx7) please open separate issue as this thread is closed. Case may or may not be the same. Thanks.
MikhailArkhipov commentedon Aug 31, 2019
Actually since it is not resolved as fixed and rather than as duplicated, reopening.
MikhailArkhipov commentedon Oct 7, 2019
anounted to torch + pandas + matplotlib + scipy + numpy, but I ended up with 1GB peak, 400M post analysis. However, this is stock P3.7
MikhailArkhipov commentedon Dec 16, 2019
LS is on 0.5, I am planning to close this unless it is still happening.