Growing node.exe memory

[nocelab]

Hi, in Windows machine I'm testing a loop with 2 load method with Playwright browser and I note that the process node.exe still grow without ever release.
Is it correct? Is there some trick to release it?

[Osiris-Team]

@nocelab can you share the code? I didnt notice anything like that when testing.

[nocelab]

this is sample code (Windows 10 with OpenJDK 8u222-b10):

HBrowser hBrowser = new HBrowser();
PlaywrightWindow hb = hBrowser.openCustomWindow().headless(false).buildPlaywrightWindow();

for (int i = 0; i < 100; i++) {
    hb.load("www.google.it");
    hb.fill("[aria-label=\"Cerca\"]", "searching something");
    hb.holdKey("Tab");
    // click and wait...
    hb.getJsContext().executeJavaScript(""+
            "await Promise.all([\n"+
            " page.click(':nth-match(:text(\"Cerca con Google\"), 2)'),\n"+
            " page.waitForNavigation()\n"+
            "]);");
    System.out.println(hb.getOuterHtml().text());
}
hb.close();

and this is memory graph of node.exe process:

[Osiris-Team]

Ok I tested this code first just to make sure node was working fine by default:

HBrowser hBrowser = new HBrowser();
        try(PlaywrightWindow window = hBrowser.openWindow()){
            window.load("www.google.com");
            System.out.println("Loaded page! Exit the program once finished with ram analysis...");
            while (true)
                Thread.sleep(1000);
        } catch (NodeJsCodeException | InterruptedException e) {
            e.printStackTrace();
        }

And it looks good since the node.exe uses only a steady amount of 31mb.

Then I tested your code (slightly modified):

HBrowser hBrowser = new HBrowser();
        try(PlaywrightWindow window = hBrowser.openCustomWindow().headless(false).buildPlaywrightWindow()){
            for (int i = 0; i < 100; i++) {
                window.load("www.google.it");
                window.fill("[aria-label=\"Cerca\"]", "searching something");
                window.holdKey("Tab");
                // click and wait...
                window.getJsContext().executeJavaScript(""+
                        "await Promise.all([\n"+
                        " page.click(':nth-match(:text(\"Cerca con Google\"), 2)'),\n"+
                        " page.waitForNavigation()\n"+
                        "]);");
                System.out.println("Run: "+i);
            }
        } catch (NodeJsCodeException e) {
            e.printStackTrace();
        }

Your code throws following exception:

com.osiris.headlessbrowser.exceptions.NodeJsCodeException: Error during JavaScript code execution! Details: 
CAUGHT JS-EXCEPTION: TimeoutError
MESSAGE: page.fill: Timeout 30000ms exceeded.
=========================== logs ===========================
waiting for selector "[aria-label="Cerca"]"
============================================================
LINE: undefined
STACK: page.fill: Timeout 30000ms exceeded.
=========================== logs ===========================
waiting for selector "[aria-label="Cerca"]"
============================================================
    at REPL12:2:12
    at REPL12:8:4


	at com.osiris.headlessbrowser.js.contexts.NodeContext.executeJavaScript(NodeContext.java:383)
	at com.osiris.headlessbrowser.js.contexts.NodeContext.executeJavaScript(NodeContext.java:304)
	at com.osiris.headlessbrowser.windows.PlaywrightWindow.fill(PlaywrightWindow.java:711)
	at com.osiris.headlessbrowser.windows.PlaywrightWindow.fill(PlaywrightWindow.java:681)
	at com.osiris.headlessbrowser.issues.Issue9.main(Issue9.java:13)

Process finished with exit code 0

Thus I wasn't able to see the same increase in memory as you since it aborts. Maybe the missing try/catch was somehow your issue?

[nocelab]

In my sample nothing goes in exception.
You have to adapt my code with your google.com right localized labels... this graph with your try/catch code:

[Osiris-Team]

Yeah was able to run your code successfully with the same results.
Looked at the java and js code of headless browser and there don't seem to be memory leaks.
I guess its playwright storing the previously loaded pages to allow back and forth navigation.

[nocelab]

Ok... I'am happy!
You are right, perhaps something like microsoft/playwright#6319.
I will deepen with Playwright... thanks!