restore: vinyl snapshot loader #7017
Conversation
ripatel-fd
force-pushed
the
ripatel/vinyl-io-wd-2
branch
2 times, most recently
from
72ab702 to
ca1e6cc
Compare
Performance Measurements ⏳
|
ripatel-fd
force-pushed
the
ripatel/vinyl-io-wd-2
branch
from
ca1e6cc to
ce878a1
Compare
Performance Measurements ⏳
|
ripatel-fd
force-pushed
the
ripatel/vinyl-io-wd-2
branch
from
ce878a1 to
63795be
Compare
Also adds additional verification of the SlotHistory sysvar
ripatel-fd
force-pushed
the
ripatel/vinyl-io-wd-2
branch
from
63795be to
1c93fc2
Compare
Performance Measurements ⏳
|
| fputs( "--------------------------------------------", stdout ); | ||
| if( snapwr_tile ) fputs( "--------------", stdout ); | ||
| fputs( "[ct],[ld],[dc],[in]--------[ct],[ld],[dc],[in]", stdout ); | ||
| if( snapwr_tile ) fputs( ",[wr]" , stdout ); |
There was a problem hiding this comment.
WDYT about similar changes in watch?
There was a problem hiding this comment.
Are the list of dummy tiles in fd_tile_unit_test.c important? (I was just looking at files I changed when adding the snapld tile)
There was a problem hiding this comment.
Not that I'm aware of since fd_tile_unit_test.c is probably not supposed to test the snapshot loader
| handle_control_frag( fd_snapwr_t * ctx, | ||
| ulong meta_ctl, | ||
| ulong meta_sig ) { | ||
| switch( meta_ctl ) { |
There was a problem hiding this comment.
does it make sense to FD_TEST that we are in IDLE state when receiving any of these?
There was a problem hiding this comment.
I don't think so, snapin->snapwr is simplified. snapin just sends a SHUTDOWN while snapin is still in PROCESSING state, because snapwr just blindly writes to offsets specified by snapin. It isn't aware what kind of snapshot is being written and whether anything was reset.
|
|
||
| populate_sock_filter_policy_fd_snapin_tile( out_cnt, out, (uint)fd_log_private_logfile_fd() ); | ||
| return sock_filter_policy_fd_snapin_tile_instr_cnt; | ||
| (void)topo; |
There was a problem hiding this comment.
OOC, are we not supposed to use FD_PARAM_UNUSED?
There was a problem hiding this comment.
Kevin prefers (void).
This patch integrates the snapshot loader with Firedancer's new disk-backed account database.
Adds a line-rate capable O_DIRECT write path with cache bypass.
Has minimal impact to the existing funk-based database logic, except for minor cleanups along the way.