On Thu, Jul 2, 2020 at 12:50 AM Vivek Goyal <vgoyal@xxxxxxxxxx> wrote:
>
> Container folks are complaining that dnf/yum issues too many sync while
> installing packages and this slows down the image build. Build
> requirement is such that they don't care if a node goes down while
> build was still going on. In that case, they will simply throw away
> unfinished layer and start new build. So they don't care about syncing
> intermediate state to the disk and hence don't want to pay the price
> associated with sync.
>
> So they are asking for mount options where they can disable sync on overlay
> mount point.
>
> They primarily seem to have two use cases.
>
> - For building images, they will mount overlay with nosync and then sync
> upper layer after unmounting overlay and reuse upper as lower for next
> layer.
>
> - For running containers, they don't seem to care about syncing upper
> layer because if node goes down, they will simply throw away upper
> layer and create a fresh one.
>
> So this patch provides two mount options "sync=off" and "sync=fs".
> First option disables all forms of sync. Now it is caller's responsibility
> to throw away upper if system crashes or shuts down and start fresh.
>
> Option "sync=fs" disables all forms of sync except syncfs/umount/remount.
> This is basically useful for image build where we want to persist upper
> layer only after operation is complete and upper will be renamed and
> reused as lower for next layer build.
>
> With sync=off, I am seeing roughly 20% speed up in my VM where I am just
> installing emacs in an image. Installation time drops from 31 seconds to
> 25 seconds when nosync option is used. This is for the case of building on top
> of an image where all packages are already cached. That way I take
> out the network operations latency out of the measurement.
>
> With sync=fs, I am seeing roughly 12% speed up.
>
> Giuseppe is also looking to cut down on number of iops done on the
> disk. He is complaining that often in cloud their VMs are throttled
> if they cross the limit. This option can help them where they reduce
> number of iops (by cutting down on frequent sync and writebacks).
>
> Changes from v2:
> - Added helper functions (Amir Goldstein)
> - Used enums to keep sync state (Amir Goldstein)
Thank you for doing that.
Sorry, but I failed to mention the practical reasons for suggesting the
enum config and you failed to read my mind :-/
So I will explain my reasoning using more words (see below)
>
> Signed-off-by: Giuseppe Scrivano <gscrivan@xxxxxxxxxx>
> Signed-off-by: Miklos Szeredi <mszeredi@xxxxxxxxxx>
> Signed-off-by: Vivek Goyal <vgoyal@xxxxxxxxxx>
Anyway, you may add:
Reviewed-by: Amir Goldstein <amir73il@xxxxxxxxx>
> diff --git a/fs/overlayfs/ovl_entry.h b/fs/overlayfs/ovl_entry.h
> index b429c80879ee..e7aea92733f5 100644
> --- a/fs/overlayfs/ovl_entry.h
> +++ b/fs/overlayfs/ovl_entry.h
> @@ -5,6 +5,14 @@
> * Copyright (C) 2016 Red Hat, Inc.
> */
>
> +enum ovl_sync_type {
> + __OVL_SYNC_OFF = (1 << 0),
> + __OVL_SYNC_FS = (1 << 1),
> +};
> +
> +#define OVL_SYNC_OFF(type) ((type) & __OVL_SYNC_OFF)
> +#define OVL_SYNC_FS(type) ((type) & __OVL_SYNC_FS)
> +
My unspoken intention when I wrote either "enum" or "flags" is -
If you use bitwise flags, they reflect the ovl_should_xxx queries:
__OVL_NOSYNC_FILE and __OVL_NOSYNC_FS
#define OVL_SYNC_FILE(type) (!((type) & __OVL_NOSYNC_FILE))
#define OVL_SYNC_FS(type) (!((type) & __OVL_NOSYNC_FS))
If you use enum (not bitwise), the distinct enum values reflect the
mount option:
OVL_SYNC_ON (=0), OVL_SYNC_OFF, OVL_SYNC_FS
I am not commenting on this because of some sort of aesthetic taste.
I am commenting on this because I think it would make parts of the
patch simpler/clearer (see below).
As far as I am concerned, for the three possible config values off/fs/on
the distinct enum values are better.
Of course, that is *my* opinion. You may disagree.
> struct ovl_config {
> char *lowerdir;
> char *upperdir;
> @@ -17,6 +25,7 @@ struct ovl_config {
> bool nfs_export;
> int xino;
> bool metacopy;
> + enum ovl_sync_type sync;
> };
>
> struct ovl_sb {
> @@ -90,6 +99,17 @@ static inline struct ovl_fs *OVL_FS(struct super_block *sb)
> return (struct ovl_fs *)sb->s_fs_info;
> }
>
> +static inline bool ovl_should_fsync(struct ovl_fs *ofs)
> +{
> + return (!OVL_SYNC_OFF(ofs->config.sync) &&
> + ! OVL_SYNC_FS(ofs->config.sync));
option #1 (bitwise):
return OVL_SYNC_FILE(ofs->config.sync);
option #2 (distinct):
return ofs->config.sync == OVL_SYNC_ON;
> +}
> +
> +static inline bool ovl_should_syncfs(struct ovl_fs *ofs)
> +{
> + return !OVL_SYNC_OFF(ofs->config.sync);
option #1 (bitwise):
return OVL_SYNC_FS(ofs->config.sync);
option #2 (distinct):
return ofs->config.sync != OVL_SYNC_OFF;
[...]
> @@ -362,6 +364,10 @@ static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
> if (ofs->config.metacopy != ovl_metacopy_def)
> seq_printf(m, ",metacopy=%s",
> ofs->config.metacopy ? "on" : "off");
> + if (OVL_SYNC_OFF(ofs->config.sync))
> + seq_puts(m, ",sync=off");
> + if (OVL_SYNC_FS(ofs->config.sync))
> + seq_puts(m, ",sync=fs");
option #1 (bitwise):
if (!ofs->config.sync)
seq_puts(m, ",sync=off");
else if (!OVL_SYNC_FILE(ofs->config.sync))
seq_puts(m, ",sync=fs");
option #2 (distinct):
Would be better. See ovl_xino_str[].
> @@ -411,6 +419,8 @@ enum {
> OPT_XINO_AUTO,
> OPT_METACOPY_ON,
> OPT_METACOPY_OFF,
> + OPT_SYNC_OFF,
> + OPT_SYNC_FS,
> OPT_ERR,
> };
>
> @@ -429,6 +439,8 @@ static const match_table_t ovl_tokens = {
> {OPT_XINO_AUTO, "xino=auto"},
> {OPT_METACOPY_ON, "metacopy=on"},
> {OPT_METACOPY_OFF, "metacopy=off"},
> + {OPT_SYNC_OFF, "sync=off"},
> + {OPT_SYNC_FS, "sync=fs"},
> {OPT_ERR, NULL}
> };
>
> @@ -573,6 +585,14 @@ static int ovl_parse_opt(char *opt, struct ovl_config *config)
> metacopy_opt = true;
> break;
>
> + case OPT_SYNC_OFF:
> + config->sync |= __OVL_SYNC_OFF;
There is really no point in ORing with old config options here.
option #1 (bitwise):
config->sync = __OVL_NOSYNC_FILE | __OVL_NOSYNC_FS;
option #2 (distinct):
config->sync = OVL_SYNC_OFF;
> + break;
> +
> + case OPT_SYNC_FS:
> + config->sync |= __OVL_SYNC_FS;
> + break;
> +
> default:
> pr_err("unrecognized mount option \"%s\" or missing value\n",
> p);
> @@ -588,6 +608,17 @@ static int ovl_parse_opt(char *opt, struct ovl_config *config)
> config->workdir = NULL;
> }
>
> + if (OVL_SYNC_OFF(config->sync) && OVL_SYNC_FS(config->sync)) {
> + pr_err("conflicting options: sync=off,sync=fs\n");
> + return -EINVAL;
> + }
> +
We are not warning user that metacopy=off conflicts with metacopy=on,
we just let the last option overwrite previous ones.
> + if (!config->upperdir &&
> + (OVL_SYNC_OFF(config->sync) || OVL_SYNC_FS(config->sync))) {
> + pr_info("option sync=off/fs is meaningless in a non-upper mount, ignoring it.\n");
> + config->sync = 0;
> + }
> +
With either bitwise/distinct options this could be simplified:
if (!config->upperdir && config->sync)
pr_info("option "sync=off/fs" is meaningless in a
non-upper mount, ignoring it.\n");
config->sync = 0;
Thanks,
Amir.
