Ruslan Dautkhanov created ZEPPELIN-2931:
-------------------------------------------

             Summary: JS memory/object leak makes Zeppelin rendering in 
browsing lag
                 Key: ZEPPELIN-2931
                 URL: https://issues.apache.org/jira/browse/ZEPPELIN-2931
             Project: Zeppelin
          Issue Type: Bug
    Affects Versions: 0.7.2, 0.7.3, 0.8.0
            Reporter: Ruslan Dautkhanov
            Priority: Critical


>From the Zeppelin users list:

{quote}
Mid-size notebooks lag in browser rendering / responding to simple navigation 
requests. 
I've seen many times that once notebooks grow up above a "small" size, 
rendering lags can happen. Running latest Chrome with a decent Intel i7 
processor. For example, a notebook I am working on has around 35 paragraphs and 
is just 145kb note.json (there are no big data tables - nothhing crazy), starts 
lagging quite a bit which doesn't help with UX.
{quote}

Response from [~pbrenner]:
{quote}
I experience this regularly. 

I have found that leaving these larger notebooks open in chrome can cause lag 
to grow worse over time. Sometimes I can bring lag back down to a manageable 
level by closing the tab completely and loading the notebook from a new tab. 
Clearing output does not seem to help, so like your colleagues I’ll sometimes 
resort to splitting to a separate notebook. 
{quote}

Further observation:
{quote}
It's interesting you mentioned that restarting a tab helps.

I just went to Chrome at it was showing that same Zeppelin tab consumes 2Gb+ of 
memory, 1Gb+ of which are JS objects. I tried to take a heap snapshot in Chrome 
devtools, but Chrome rendering engine crashed so had to restart that tab. After 
tab was restarted, that tab's heap consumption went down to just 82Mb - and it 
renders now without lags! It looks to me some sort of JS objects / JS memory 
leak. 
{quote}

After I restarted tab and left it running for 12 hours, memory consumption of 
the Tab went up to 400Mb, 122Mb of which are JS related:


It may take a few days to go back to 2Gb/1Gb as it was before.




--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Reply via email to