pmem / redis Goto Github PK

createStringObjectFromLongLongA - there should be a memkind allocation for embedded string/int

Hi!
I would like to use REDIS on a centos 7 machine with only PMEM but unfortunately I cannot build the software.
Here are the commands that I run:

yum -y install memkind-devel
git clone https://github.com/pmem/redis.git && cd redis/
make MALLOC=memkind

Here are the errors:
error: ‘MEMKIND_DAX_KMEM’ undeclared
error: ‘MEMKIND_STAT_TYPE_RESIDENT’ undeclared
error: ‘MEMKIND_STAT_TYPE_ACTIVE’ undeclared
error: ‘MEMKIND_STAT_TYPE_ALLOCATED’ undeclared

I also tried with the following make command and I got the same errors:
make USE_PMDK=yes STD=-std=gnu99 MALLOC=memkind

Any suggestions?

Thanks a lot

Having myconfig.conf file with following registration:

pmfile /mnt/non-existent-dir/redis.pm 3gb

and starting redis-server:
./redis-server myconfig.conf
we get:

760:C 07 Feb 18:02:52.638 * Start init Persistent memory file /mnt/non-existent-dir/redis.pm size 3.00G
Segmentation fault (core dumped)

The same situation is when starting two processes of redis-server with two config files pointing to the same pmfile path (which exists in this case)
Redis version: 3.2.7-25-g1e7aa59
Pmdk version: 1.3+b2-614-g1bc6897

[root@shz23f_aep_nc04 redis]# make USE_PMDK=yes STD=-std=gnu99
cd src && make all
make[1]: Entering directory '/root/yazhan/redis/src'
CC dict.o
In file included from dict.c:49:0:
server.h:744:29: error: ‘POBJ_MAX_ACTIONS_TYPE’ undeclared here (not in a function); did you mean ‘POBJ_MAX_ACTION_TYPE’?
struct pobj_action actions[POBJ_MAX_ACTIONS_TYPE];
^~~~~~~~~~~~~~~~~~~~~
POBJ_MAX_ACTION_TYPE
dict.c: In function ‘dictReplacePM’:

I am executing with make test the redis tests inside a CentOS Stream VM, all the others passed but the bitfield one fails with the following messages:

!!! WARNING The following tests failed:

*** [err]: BITFIELD overflow detection fuzzing in tests/unit/bitfield.tcl
OW not detected where needed: i64 -9223372036854775808+-9223372036854775808
*** [err]: BITFIELD overflow wrap fuzzing in tests/unit/bitfield.tcl
WRAP error: i64 -9223372036854775808+-9223372036854775808 = 9223372036854775807, should be 0
Cleanup: may take some time... OK
make[1]: *** [Makefile:327: test] Error 1
make[1]: Leaving directory '/home/redis/src'
make: *** [Makefile:6: test] Error 2

I configured emulation version of pmem referring official blog.
link

And than, build & configure pmem/redis to use pmem.

# Build pmem/redis
$ make USE_PMDK=yes STD=-std=gnu99

# Set pmfile to our mounted pmem(emulation) in redis.conf
$ vim redis.conf
# pmfile /path/to/pmem-mount/redis.pm 1gb

...and run redis-server by this command.

$ ./src/redis-server redis.conf

then, server is executed well like this.

I execute redis-benchmark for pmem/redis. Then I could see that SET operation has worse performance than GET operation.

evaluated time to complete 100000 SET operation: 16.72 sec
evaluated time to complete 100000 GET operation: 0.63 sec

This is result of redis-benchmark on pmem/redis without using pmem.

evaluated time to complete 100000 SET operation: 0.63 sec
evaluated time to complete 100000 GET operation: 0.63 sec

Without pmem, benchmark show that evaluated time to operation is same both GET and SET.
I just wonder why SET has worse performance than GET on pmem.
Is there any additional configuration for pmem/redis? (ex. CPU setup, cache setup, ...)

• If the pool is corrupted or is otherwise unopenable, pmemobj_open(...) will return NULL, causing a SEGFAULT during initialization (i.e., at POBJ_ROOT).

Hi experts,

The current redis-nvml is based on v3.2 and it does not support the multithreading I/O that is the new feature introduced since redis 6 (e.g., 1. https://www.i-programmer.info/news/197-data-mining/13682-redis-6-adds-multithreading-io.html;
2. https://www.infoworld.com/article/3541356/redis-6-arrives-with-multithreading-for-faster-io.html#:~:text=With%20I%2FO%20threading%20enabled,a%20single%20instance%2C%20say%20creators&text=Redis%206%2C%20the%20newest%20version,moved%20to%20general%20availability%20today.)

Is there any open-source redis-nvml version that supports the multithreading I/O feature? Or is there any ongoing plan to support the multithreading I/O feature in redis-nvml?

Thanks!

Hi!

I'm trying to use only the PMEMs but when I change the configuration file (redis.conf) to "only-pmem" at line 1788 I get the following error. Have you encountered this issue before?

I built the project on a centos 8 machine using the memkind allocator.

Thanks
Adam

=== REDIS BUG REPORT START: Cut & paste starting from here ===
191495:M 16 Jul 2020 11:38:21.300 # Redis 5.9.102 crashed by signal: 11
191495:M 16 Jul 2020 11:38:21.300 # Crashed running the instruction at: 0x7ff622b8cb1d
191495:M 16 Jul 2020 11:38:21.300 # Accessing address: (nil)
191495:M 16 Jul 2020 11:38:21.300 # Failed assertion: (:0)

------ STACK TRACE ------
EIP:
/lib64/libc.so.6(vsnprintf+0x7d)[0x7ff622b8cb1d]

Backtrace:
./redis-server 127.0.0.1:6379(logStackTrace+0x2e)[0x46e93e]
./redis-server 127.0.0.1:6379(sigsegvHandler+0x96)[0x46efb6]
/lib64/libpthread.so.0(+0x12dd0)[0x7ff622ee7dd0]
/lib64/libc.so.6(vsnprintf+0x7d)[0x7ff622b8cb1d]
/lib64/libc.so.6(__snprintf+0x93)[0x7ff622b6c2e3]
./redis-server 127.0.0.1:6379(redisAsciiArt+0xc9)[0x431ac9]
./redis-server 127.0.0.1:6379(main+0x3c6)[0x4254f6]
/lib64/libc.so.6(__libc_start_main+0xf3)[0x7ff622b366a3]
./redis-server 127.0.0.1:6379(_start+0x2e)[0x42583e]

------ INFO OUTPUT ------
Segmentation fault (core dumped)

should s/USE_MEMKING/USE_MEMKIND ?
https://github.com/pmem/redis/blob/5.0-poc_cow/src/object.c#L517

libjemalloc should be linked from jemalloc, not memkind. This can cause some problems in statistics, stability etc.

src/Makefile:204
$(MEMKIND_PATH)/.libs/libmemkind.a $(MEMKIND_PATH)/jemalloc/obj/lib/libjemalloc.a

Investigate places, described by:
#144 (review)
#147 (review)
#147 (review)