Hrm, this looks like it could be the cause:

https://cs.chromium.org/chromium/src/content/browser/frame_host/render_frame_message_filter.cc?sq=package:chromium&l=460

https://cs.chromium.org/chromium/src/content/browser/frame_host/render_frame_message_filter.cc?sq=package:chromium&l=492

"Pass a null callback since we don't care about when the 'set' completes."

------

A couple runs, with some key methods DLOGing, where 'A' is the cookieStore.delete(), 'B' is the set via document.cookie, and 'C' is the cookieStore.get()

We're dealing with two processes (renderer, browser) and the browser has two threads, so logging can be confusing.



Succeeded:

A blink::ScriptPromise blink::CookieStore::DoWrite()
A - void net::CookieMonster::SetCanonicalCookie()
A static void blink::CookieStore::OnSetCanonicalCookieResult()

B void blink::Document::setCookie()
B void blink::SetCookies()

C blink::ScriptPromise blink::CookieStore::DoRead() (**1)
B - virtual void content::RenderFrameMessageFilter::SetCookie() - invoking callback (**2)
B - virtual void content::RenderFrameMessageFilter::SetCookie() - issuing async set
B - void net::CookieMonster::SetCanonicalCookie() (**3)
C - virtual void net::CookieMonster::GetCookieListWithOptionsAsync()
C - void net::CookieMonster::GetCookieListWithOptions() (**4)
C static void blink::CookieStore::GetAllForUrlToGetResult()

Note that log ordering of **1 and **2 is presumably due to multi-thread/process logging; the DoRead call (**1) should not happen until after the callback runs (**2). 

The important thing is that the Set (**3) happens before the Get (**4), on the CookieMonster's thread.

....

Failed:

A blink::ScriptPromise blink::CookieStore::DoWrite()
A - void net::CookieMonster::SetCanonicalCookie()
A static void blink::CookieStore::OnSetCanonicalCookieResult()

B void blink::Document::setCookie()
B void blink::SetCookies()

C blink::ScriptPromise blink::CookieStore::DoRead()
C - virtual void net::CookieMonster::GetCookieListWithOptionsAsync()
C - void net::CookieMonster::GetCookieListWithOptions() (**1)
B - virtual void content::RenderFrameMessageFilter::SetCookie() - invoking callback
B - virtual void content::RenderFrameMessageFilter::SetCookie() - issuing async set
B - void net::CookieMonster::SetCanonicalCookie() (**2)
C static void blink::CookieStore::GetAllForUrlToGetResult()

Note here that the Get (**1) happens before the Set (**2), on the CookieMonster's thread. :(

Hrm, that may not be the (only) issue. If I pass the callback through so it is invoked after the set is (theoretically) complete, but back out all the logging (*sigh*) the the flakiness is unchanged.