[CRIU] [PATCH] pagemap: fix reading pages from socket for --remote case

[Pavel Emelyanov]

On 04/25/2017 10:54 AM, Rodrigo Bruno wrote:
> Hi,
> 
> 2017-04-24 12:55 GMT+01:00 Pavel Emelyanov <xemul at virtuozzo.com <mailto:xemul at virtuozzo.com>>:
> 
>     On 04/23/2017 04:29 AM, Rodrigo Bruno wrote:
>     > Hi,
>     >
>     > the patch looks good. It avoids using 'pread' which is not supported for remote images.
>     >
>     > However, the patch that I sent for the mailing list introducing remote images used 'read' instead of 'pread'.
> 
>     Yes, sorry about that, it was me wrongly merging async reads with remotes :(
> 
>     I have a question regarding this place, while we're at it. Here's the code from
>     maybe_read_page_local():
> 
>             if ((flags & (PR_ASYNC|PR_ASAP)) == PR_ASYNC && !opts.remote)
>                     ret = enqueue_async_page(pr, vaddr, len, buf);
>             else {
>                     ret = read_local_page(pr, vaddr, len, buf);
> 
>     Why don't we enqueue the read request for opts.remote case?
> 
> 
> To be honest, I don't remember why I avoided async pages... I looked through the code and I don't see any
> particular reason for it not to work with remote images.
> 
> Anyway, just to clarify, what is the motivation behind async pages? Are we avoiding any page read that might be unnecessary
> in the future? 

No, for on-disk files doing many small read()-s is still slow even if the data
is in cache. With async pages we merge all the reads together and then call
single preadv() doing all IO with one call.

-- Pavel