cuju-ft / cuju Goto Github PK

I tried to install Cuju in a vmware virtual machine. This error occurred when I was running the QEMU image. Is there a solution?
W: /etc/qemu-ifup: no bridge for guest interface found
kvm_set_phys_mem: error registering slot: Inappropriate ioctl for device
Aborted (core dumped)

qemu-system-x86_64: migration/migration.c:2906: migrate_timer: Assertion `!kvm_shm_clear_dirty_bitmap(s->cur_off)' failed.

I think it need to make it re-sync

In KVM module, the related dirty bitmaps should be released include epoch_dirty_bitmaps. Although in the commit b48e0b7 temporarily fix the issue of the kernel panic due to the destruction of related epoch_dirty_bitmaps, just ignore and doesn't really release them.

I would like to ask what configuration scenarios I need to build for Centos7 operating system at compile time.Could you please provide the document for reference? Thank you.

Hello, I would like to ask whether these codes are still being updated? Virtual machine memory synchronization is a good topic, can you use libvirt to manage virtual machine management?

When failover , If we call virtio_blk_get_request_from_head to load record_list request, it will load wrong request and cause the assert(type & VIRTIO_BLK_T_OUT) check fail.

Now we find some fix to solve this problem:

1. Change the method of virtqueue_pop get request address (no use undo_map)
2. Add get req->sector_num function on virtio_blk_get_request_from_head()
3. Defer virtio_blk_free_request() from virtio_blk_rw_complete() to virtio_blk_complete_head()

And we still find virtio_blk_get_request_from_head sometimes load a wrong request (type !=1)
now maybe the problem is some dirty page produce from qemu and not be tracked (cpu_physical_memory_map)
the defer_write_map mechanism maybe is wrong

When start FT mode, if we use

sudo nc -U /mnt/vm1.monitor

to enter qemu monitor
It will cause sg

Backtrace:

#0  _int_malloc (av=av@entry=0x7ffff6ab7b20 <main_arena>, bytes=bytes@entry=65) at malloc.c:3516
#1  0x00007ffff67762b0 in _int_realloc (av=av@entry=0x7ffff6ab7b20 <main_arena>, oldp=oldp@entry=0x1637e60, oldsize=oldsize@entry=48, nb=nb@entry=80)
    at malloc.c:4304
#2  0x00007ffff6777839 in __GI___libc_realloc (oldmem=0x1637e70, bytes=63) at malloc.c:3045
#3  0x00007ffff74507d8 in g_realloc () from /lib/x86_64-linux-gnu/libglib-2.0.so.0
#4  0x0000000000742a43 in capacity_increase (qstring=qstring@entry=0x1626320, len=len@entry=1) at qobject/qstring.c:75
#5  0x0000000000742bc5 in qstring_append_chr (qstring=0x1626320, c=112) at qobject/qstring.c:104
#6  0x000000000044c406 in monitor_puts (mon=mon@entry=0x1667580, str=0x7fff040ef28f "' for more information\n", 
    str@entry=0x7fff040ef270 "QEMU 2.8.0 monitor - type 'help' for more information\n") at /mnt/nfs/Cuju/monitor.c:339
#7  0x000000000044cfa2 in monitor_vprintf (mon=0x1667580, fmt=<optimized out>, ap=<optimized out>) at /mnt/nfs/Cuju/monitor.c:359
#8  0x000000000044d057 in monitor_vprintf (ap=0x7fffffffdd30, fmt=0x76c678 "QEMU %s monitor - type 'help' for more information\n", mon=<optimized out>)
    at /mnt/nfs/Cuju/monitor.c:354
#9  monitor_printf (mon=<optimized out>, fmt=fmt@entry=0x76c678 "QEMU %s monitor - type 'help' for more information\n") at /mnt/nfs/Cuju/monitor.c:367
#10 0x000000000044ee56 in monitor_event (opaque=0x1667580, event=<optimized out>) at /mnt/nfs/Cuju/monitor.c:3905
#11 0x00000000005311c3 in tcp_chr_new_client (chr=chr@entry=0x1617870, sioc=sioc@entry=0x7fff040ef020) at qemu-char.c:3373
#12 0x0000000000531290 in tcp_chr_accept (channel=<optimized out>, cond=<optimized out>, opaque=0x1617870) at qemu-char.c:3409
#13 0x00007ffff744b04a in g_main_context_dispatch () from /lib/x86_64-linux-gnu/libglib-2.0.so.0
#14 0x00000000006b42d5 in glib_pollfds_poll () at main-loop.c:220
#15 os_host_main_loop_wait (timeout=<optimized out>) at main-loop.c:265
#16 main_loop_wait (nonblocking=<optimized out>) at main-loop.c:513
#17 0x0000000000410fbc in main_loop () at vl.c:1984
#18 main (argc=<optimized out>, argv=<optimized out>, envp=<optimized out>) at vl.c:4725

savevm.c:2799: qemu_savevm_trans_complete_precopy_advanced: Assertion `ftdev->state_entry_num < CUJU_FT_DEV_STATE_ENTRY_SIZE' failed.

CUJU_FT_DEV_STATE_ENTRY_SIZE in include/migration/migration.h
#define CUJU_FT_DEV_STATE_ENTRY_SIZE 50

If we want to use -smp 32
We need change to CUJU_FT_DEV_STATE_ENTRY_SIZE 150
Maybe it need a dynamic method to handle this thing

In fine-grained dirty region tracking optimization
Some diff page size is 0, it means three possibilities:

1. Page content is from A to B and change back to A
2. Page content is not dirty but we mark it dirty
3. Backup too late

If in case 1 or 2, we can ignore it.
In case 3, we will transfer all page, otherwise it will missing logging some dirty page.

But the correct solution is find what dirty page we backup too late
This issue is correspond to 60623ee

1. add sudo in command if working on /mnt/ directory
2. remind user to create a directory /home/[username]/nfsfolder before exit /etc/exports in NFS Node Setup
3. mark all path names as "Code" to increase readability.
   Thanks!

Hi,

I'm new to VM fault tolerance. I'm not sure if Cuju guarantees the shared storage state or not. Some fault-tolerance solutions use either storage snapshot, or a private storage for the secondary VM, to preserve the storage state while checkpointing. How does Cuju handle this issue? Alternatively, is there any best practice for using shared storage? Thanks.

When open vim and type fast, the screen may get stuck if cpu setting more than one.
there are two cases below :
-cpu host (the screen would not get stuck)
-cpu host -smp <NUMBER> (if NUMBER more than one the screen would get stuck easily)

memslot = gfn_to_memslot(kvm, gfn);
if (unlikely(!memslot)) {
printk(KERN_ERR"%s can't find memslot for %lx\n", func, gfn);
memslots_dump(kvm);
return -ENOENT;
}

Can not use gfn find memslot if gfn > DAC00

This guest image is not change to text mode ubuntu 16.04 guest image
https://drive.google.com/file/d/1JQrJRw48nC8OJqFIwa6AU9eDBeYGQ39j/
When we use this guest image and enter FT mode, vncviewer can not update display.