Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    John Sully
    @JohnSully
    @Yahook Performance of the current FLASH feature drops off pretty quickly as the dataset exceeds RAM. We have a new version of FLASH coming out for January that is much faster.
    hezhiqiang
    @hezhiqiang
    can keydb be a slave of redis?
    hezhiqiang
    @hezhiqiang
    when keydb is active-active mode, which one should redisson user: replicated-servers, master-slave-servers, cluster-servers?
    hezhiqiang
    @hezhiqiang
    when i make keydb in centos, show this error: unrecognized command line option ‘-std=c++14’
    hezhiqiang
    @hezhiqiang
    i installed gcc 4.9.2, std=c++14 error is solved, but now is uuid/uuid.h: No such file or directory
    hezhiqiang
    @hezhiqiang
    now is this error: compactvector.h:18:35: error: ‘is_trivially_copyable’ is not a member of ‘std’
    static_assert(MEMMOVE_SAFE || std::is_trivially_copyable<T>::value, "compactvector requires trivially copyable types");
    which version of gcc shoud i use?
    hezhiqiang
    @hezhiqiang
    now make is sucessful, but when make test, has this error:
    * [err]: BITOP lshift string in tests/unit/bitops.tcl
    Expected 'ïg' to equal or match 'abcdefg'
    hezhiqiang
    @hezhiqiang
    haha, i installed ok, but when i set keydb be a slave of redis, has this error: non-fatal: Master doesn't understand REPLCONF uuid
    张舒鳌
    @zhangshuao
    hi gays
    @hezhiqiang yum -y install uuid-devel can resolve it .(i installed gcc 4.9.2, std=c++14 error is solved, but now is uuid/uuid.h: No such file or directory)
    张舒鳌
    @zhangshuao
    @JohnSully Can write an installation document about CentOS 7 in Wiki ?
    张舒鳌
    @zhangshuao
    make MALLOC=memkind
    John Sully
    @JohnSully
    Yes KeyDB can replicate from Redis
    sorry about the compile issues your having. Almost everyone is pulling from docker these days so thats what we focus on.
    Please open a bug to track a CentOS build instructions document and I’ll get to it when I have a chance
    张舒鳌
    @zhangshuao
    @JohnSully This is my issue: JohnSully/KeyDB#118
    make MALLOC=memkind (flash support)
    lt1946
    @lt1946
    cluster keydb ,can use pipline or multi commend?
    John Sully
    @JohnSully
    Yes. If we don’t have 100% feature parity with Redis its a bug.
    lt1946
    @lt1946
    can u prove , I run at springboot has errors ,same as redis5, not use multi commend,
    John Sully
    @JohnSully
    Can you give me repro steps? I’m not familiar with springboot. The multi command does work as far as I’ve been able to test.
    Heng Kuang
    @hengku
    @JohnSully will RDB save and AOF rewrite benefit from multi-threaded keydb server?
    Sayan Nandan
    @sntdevco
    Hey, when are you planning to add information on a multi-master set up?
    Sayan Nandan
    @sntdevco
    Oops, looks like it's already there
    John Sully
    @JohnSully
    @hengku I have plans to improve perf there but at the moment RDB is not any faster. There may be some benefits to AOF but I haven’t done any benchmarks.
    I do have ideas on how to make this faster in the future though.
    Sayan Nandan
    @sntdevco
    @JohnSully What makes multi master unstable?
    John Sully
    @JohnSully
    @sntdevco it suffers from network traffic amplification on the order of n log n. This may not be a problem if your cluster size is small or you don’t have a lot of writes.
    Sayan Nandan
    @sntdevco
    Hmmm okay
    Does redis cluster have the same issue?
    John Sully
    @JohnSully
    Redis clusters only contain a subset of the keyspace so there isn’t a need to send data to other nodes. Active Replicas have a full copy of the database
    A lot of people are using multi master successfully but I don’t want to call it stable until i bring the network traffic down.
    KeyDB also supports traditional Redis clustering as well if that works better for your use case.
    hezhiqiang
    @hezhiqiang
    how to upgrade KeyDB without stopping service?
    Numeriku
    @Numeriku
    so to use this as a replacement for my single redis instance, i can just stop redis and compile keydb to replace my redis instance? How do i make it read my existing redis.conf when compiling it?
    hezhiqiang
    @hezhiqiang
    i found a bug, i use LPush to push 17786 records, but keydb is 88868 records, i have two server with active replication, and i use haproxy.
    pavman42
    @pavman42

    I'm attempting to replace our nodejs <-> redis production config with keydb as we have a need for multi-master syncing across datacenters. We're running 2x docker swarm currently with 1 redis replica per site via docker-ce (so the swarms act independently and don't know about each other).

    Question: Is the only way to do this --replicaof=<opposite potential nodes> or is there a better way to do this?
    Trying to prevent a single point of failure here without having to specify every potential node in every site or having to rearchitect this with haproxy or keepalived.

    Appreciate any insights into this as it seemd to work, but the "unavailable" replicas spam the logs with: Unable to connect to MASTER: Resource temporarily unavailable

    Essentially, does this add any unnecessary overhead by specifying unavailable masters that might hamper the performance if it remains in this state? If not, is there anyway to suppress this output? :)

    John Sully
    @JohnSully
    Hi @pavman42 autodiscovery has been a feature request of @nagashev for quite a while. It's on my list of things to do eventually. But for now yes you must specify explicitly all nodes you want to connect to.
    That log is at warning level so not a good way to silence it. I think what I'll do is drop the severity to info if multimaster is enabled. That way you can silence it by changing the log level
    John Sully
    @JohnSully
    Change edb993d5 drops the severity of the master disconnect warning to INFO
    to filter out those messages you can add "loglevel warning" to your configuration. The change will be in the next release
    Teoti Nathaniel
    @teotikalki
    Is there an example of a systemd setup? The Debian installer only seems to provide initd and I've been unable to make it work for some reason (ie using known working sytemd setups from other things and commands that work from the ci)
    pavman42
    @pavman42

    I think I had a misconfiguration in ansible, as it stopped spamming once I fixed the right host groups and redeployed.

    I did notice that if I didn't set both the master password & "default" password the same, I was getting Unable to AUTH to MASTER: -WRONGPASS invalid username-password pair but that could be again due to the previous misconfiguration with ansible. I haven't gone back to try different passwords; if it's required you may want to make that clearer in the example as I had tried two different credentials and was getting that error. I'll test in a few just to be sure, probably pebkac.

    Appreciate the quick responses!

    pavman42
    @pavman42
    Circling back, looks like synchronized passwords is a feature. Not sure if this is normal behavior, but running --masterauth blah w/ a --masterpassword foo and a --requirepass bar appears to break things in multi-master. No biggie since we have a pretty simple setup, but would be nice to have different default credentials, assuming nothing slows down. I have not tried setting a specific non-default user to see if there is a difference, but it's not worth messing around with right now. Thanks again!
    kevinsteger
    @kevinsteger
    The Coming Soon states "Using Multi-Master Feature" yet the multiple masters page says "KeyDB now has support with multiple masters ...". So, is this feature available today? The docs are contradicting.
    John Sully
    @JohnSully
    @kevinsteger multimaster is available and we have quite a few users. The reason the docs are this way is that it’s still considered an experimental feature.
    Two reasons for that. 1) There can be excessive network traffic with large clusters. And 2) this bug: JohnSully/KeyDB#111
    tuapuikia
    @tuapuikia
    Hi @JohnSully , I am promoting keydb in my company as we know the different between redis and keydb is the ACID and active-replication features.
    do we have atomicity if we configure the cluster to use active-replication to accept command from all the nodes?