In my short observation varying the speed, the latest request which is in-flight when checking the latest script looks fired again.

horo-san: could you take a look?

I think this is a problem of the way how HTMLResourcePreloader and InspectorNetworkAgent use MemoryCache.

1. When there is two <script> tags (without async), "second.js" is preloaded by
   HTMLResourcePreloader::preload(). And the script resource is stored in the
   MemoryCache.
    Example:
     <script src="first.js"></script>
     <script src="second.js"></script>

2. When Chrome finished loading (and executing) "first.js",
   HTMLScriptRunner::execute() takes "second.js" from the MemoryCache and execute.

3. One second after the page is loaded, Chrome starts a new ServiceWorker to
   check the script update.

4. While starting the new ServiceWorker, MemoryCache is cleared by
   MemoryCache::evictResources() twice.
    1. InspectorNetworkAgent::setCacheDisabled() calls MemoryCache::evictResources()
       while attaching the DevTools to the ServiceWorker.
    2. InspectorNetworkAgent::didCommitLoad() calls MemoryCache::evictResources()
       when the shadow page for the ServiceWorker is loaded by
       WebEmbeddedWorkerImpl::loadShadowPage()

If MemoryCache::evictResources() is called while loading (and executing) "first.js",
"second.js" is loaded from the server via the ServiceWorker again.

I can reproduce this issue without ServiceWorker.
https://youtu.be/OLACsElqIUU

pfeldman@
I think this is an issue of "Disable cache" check box in DevTools.
Could you please handle this?

 Issue 657914  has been merged into this issue.

I am only able to reproduce the bug in my case (https://bugs.chromium.org/p/chromium/issues/detail?id=657914) only when both service worker and disable cache are enabled. Duplicate requests don't happen when I check the bypass for network option. 

If I open Chrome with  open /Applications/Google\ Chrome\ Canary.app --args --disk-cache-dir=/dev/null --disk-cache-size=1 --media-cache-size=1
I am not able to reproduce the bug..

cbentzel@, japhet@: I'm looking for someone to own disabled cache semantics as a service, in a way that is compatible with all network features (h2, service workers, etc). Who should own this?

@horo: could you triage this?