test,win: speedup tls-server-verify - #1836
Conversation
Sorry, something went wrong.
|
cc @nodejs/crypto |
Sorry, something went wrong.
|
<3 |
Sorry, something went wrong.
|
what are the chances of getting these upstreamed to openssl? |
Sorry, something went wrong.
|
@rvagg No chance because this makes a client connection be in a low entropy state. This change only applied to openssl s_client command on Windows and no affects to libopenssl.a so that iojs has a good entropy state. |
Sorry, something went wrong.
|
@shigeki Why is it a low entropy? RAND_poll(CryptGenRandom) still gets called and for the CI the screen content is always the same, so not much entropy added in this case. |
Sorry, something went wrong.
There was a problem hiding this comment.
Nitpick: name it -no_rand_screen for consistency with other options.
Sorry, something went wrong.
There was a problem hiding this comment.
I agree it. Fixed.
Sorry, something went wrong.
|
@mathiask88 I should have written it as "lower than the current". I don't think it is absolutely low for RAND_add with CryptGenRandom but I can't say it is enough high without RAND_screen for general uses. In CI, I agree that we need not it for this server-verify tests. |
Sorry, something went wrong.
|
@orangemocha According to Ben's comments, my commits were updated in shigeki@833b236 , shigeki@9f0f7c3 and shigeki@2ff517e . |
Sorry, something went wrong.
OpenSSL s_client introduces some delay on Windows. With all clients running sequentially, this delay is big enough to break CI. This fix runs the clients in parallel (unless the test includes renegotiation), reducing the total run time. (cherry picked from commit 12d4ea3c8ae7f8011de9ca9f50bdc08ee8cfb8bf)
Different servers must use different ports. Since we can count only on common.PORT and common.PORT+1, run only 2 servers in parallel. (cherry picked from commit efb0075ee0e1f4123974fe5b74ded8727bad4a9e)
When running in parallel, it is not easy to identify what server and client failed when the test fails. This adds identifiers to all lines of console output. (cherry picked from commit 29c651522df2f47c4360660264eb857fb6232b23)
For better performance of the test, the parent kills child processes so as not to wait them to be ended. (cherry picked from commit 833b23636045f7afc929196139021630a390391a)
In openssl s_client on Windows, RAND_screen() is invoked to initialize random state but it takes several seconds in each connection. This added -no_rand_screen to openssl s_client on Windows to skip RAND_screen() and gets a better performance in the unit test of test-tls-server-verify. Do not enable this except to use in the unit test. (cherry picked from commit 9f0f7c38e6df975dd39735d0e9ef968076369c74)
This improves the performance of openssl s_client on Windows and gains several seconds to finish test-tls-server-verify. (cherry picked from commit 2ff517e0e410ea33ba5a3d289a82fc315d120e8e)
|
@bnoordhuis Thanks for your comments, I updated my commits. I also picked the updated commits from @shigeki . |
Sorry, something went wrong.
Sorry, something went wrong.
|
Thanks, @bnoordhuis . I will land this shortly. |
Sorry, something went wrong.
OpenSSL s_client introduces some delay on Windows. With all clients running sequentially, this delay is big enough to break CI. This fix runs the clients in parallel (unless the test includes renegotiation), reducing the total run time. Fixes: nodejs#1461 PR-URL: nodejs#1836 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Different servers must use different ports. Since we can count only on common.PORT and common.PORT+1, run only 2 servers in parallel. Fixes: nodejs#1461 PR-URL: nodejs#1836 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
In openssl s_client on Windows, RAND_screen() is invoked to initialize random state but it takes several seconds in each connection. This added -no_rand_screen to openssl s_client on Windows to skip RAND_screen() and gets a better performance in the unit test of test-tls-server-verify. Do not enable this except to use in the unit test. Fixes: nodejs#1461 PR-URL: nodejs#1836 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
In openssl s_client on Windows, RAND_screen() is invoked to initialize random state but it takes several seconds in each connection. This added -no_rand_screen to openssl s_client on Windows to skip RAND_screen() and gets a better performance in the unit test of test-tls-server-verify. Do not enable this except to use in the unit test. Fixes: nodejs#1461 PR-URL: nodejs#1836 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
In openssl s_client on Windows, RAND_screen() is invoked to initialize random state but it takes several seconds in each connection. This added -no_rand_screen to openssl s_client on Windows to skip RAND_screen() and gets a better performance in the unit test of test-tls-server-verify. Do not enable this except to use in the unit test. Fixes: nodejs#1461 PR-URL: nodejs#1836 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
In openssl s_client on Windows, RAND_screen() is invoked to initialize random state but it takes several seconds in each connection. This added -no_rand_screen to openssl s_client on Windows to skip RAND_screen() and gets a better performance in the unit test of test-tls-server-verify. Do not enable this except to use in the unit test. Fixes: nodejs#1461 PR-URL: nodejs#1836 Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Addresses #1461 (and nodejs/node-v0.x-archive#25279 once ported there).
With the -no-rand-screen changes, the test is now much much faster (less than 1s on my machine). I am not sure however if speeding up a test is a good justification for adding a floating patch in openssl. I added all the commits in here to get more feedback.