will leverage go template to not affect performance of application, an agent pull kv and transfer it into a file that app can understand
wjm wjm <zzz...@gmail.com> 于2019年4月4日周四 下午4:41写道： > do we have a mechanism to use the configuration with high performance？ > > Xiaoliang Tian <xiaoliang.t...@gmail.com> 于2019年3月29日周五 下午5:27写道： > > > sure, mainly forcus on management , ease of use, rollback, key value > > version control. value validation. > > > > first we support API, later UX. > > > > bismy <bi...@qq.com> 于2019年3月29日周五 下午4:08写道： > > > > > Provide a centered configuration management service is very important > for > > > ServiceComb, because all microservice governance functions are based on > > > dynamic properties. In ServiceComb microservice view, now we currently > > use > > > following functions( implement archaius api using apollo and > > config-center) > > > > > > > > > * pull/push configurations of key-value pair from configuration > > service. > > > > > > > > > in microservice view, no future functions is needed. So I guess the > > > suggested service mainly focus on management functions. How do you > decide > > > to provide this functions? Add a management console or by REST api? > > > > > > > > > ------------------ 原始邮件 ------------------ > > > 发件人: "Xiaoliang Tian"<xiaoliang.t...@gmail.com>; > > > 发送时间: 2019年3月29日(星期五) 下午2:14 > > > 收件人: "dev"<dev@servicecomb.apache.org>; > > > > > > 主题: Re: [PROPOSAL] ServiceComb MetaConfig > > > > > > > > > > > > let's think how we change a timeout for a particular operation? > > > the key is xxx.usermanagement.user.getuser.timeout, is that key easy to > > be > > > manged by operator? I guess he has to learn a special key rule only > make > > > sense in his team > > > > > > I want this timeout just be timeout, and easy to read > > > it bacame timeout(service= usermanagement , > > schema=user,opertion=getuser) > > > > > > Xiaoliang Tian <xiaoliang.t...@gmail.com> 于2019年3月29日周五 下午12:15写道： > > > > > > > > > > > maybe the name confused，it is a service only for configurations，but > the > > > key contains metadata > > > > key1: log_level(service=cart, version=1.0, app=mall) > > > > key2: log_level(service=cart, version=1.0, app=mall, ip=x.x.x.x) > > > > that is 2 different keys > > > > > > > > you can consider apollo has fixed metadata and one fixed view to math > > > > their own scenario. > > > > > > > > but in apollo, key can be very complicated, user can not easily > > observe > > > > keys, a lot of metadata will be attacted to key name, just like ip to > > > > "log_level" > > > > > > > > Willem Jiang <willem.ji...@gmail.com> 于2019年3月29日周五 上午11:36写道： > > > > > > > >> Hi Xiaoliang, > > > >> > > > >> Do you propose we start a new project to manage the micro services > > meta > > > >> datas? > > > >> Currently we just have a solution to integrate with with Apollo for > > > >> configuration management, but what's the difference between the > > > >> service metadata and configuration? > > > >> I could help us to understand the proposal, if we can drop a clear > > line > > > >> for it. > > > >> > > > >> > > > >> Willem Jiang > > > >> > > > >> Twitter: willemjiang > > > >> Weibo: 姜宁willem > > > >> > > > >> On Fri, Mar 29, 2019 at 10:52 AM Xiaoliang Tian > > > >> <xiaoliang.t...@gmail.com> wrote: > > > >> > > > > >> > MetaConfig is a service for configuration management in an large > > > >> scaled > > > >> > distrbuted system. > > > >> > > > > >> > Currently we use ctrip appllo for config managment, but it is > build > > > for > > > >> a > > > >> > paticular scenario. it has fixed schema > > > >> application,environment,cluster and > > > >> > namespace. Each key belong to that schema. > > > >> > > > > >> > But it is not enough for operator to manage a complicated > > > >> infrasctructure > > > >> > or application cluster. > > > >> > > > > >> > So here is the proposal to build a brand new config management > > > service, > > > >> > > > > >> > *Features: * > > > >> > *key with rich metadata: *user can define metadata for a key , > that > > > >> > distinguish from key to another key. a key will not be stringed > by > > > >> fixed > > > >> > schema. metadata is like "env=test, service=cart, version=1.0" or > > > >> > "cluster=xxx" or "env=test, service=cart, version=1.0, > ip=x.x.x.x" > > > >> > *validator: *value can be checked by user defined python script, > so > > in > > > >> > runtine if someone want to change this value, the script will > check > > if > > > >> this > > > >> > value is properate. > > > >> > *encryption webhook: *. value can by ecrypt by your custom > > encryption > > > >> > service. > > > >> > *event producer: *producer will send key changes event to a > target > > > >> > service(kafka,rabbitmq, customed). > > > >> > *config view: *by setting metadata criteria, MetaConfig will > > > aggregate a > > > >> > view to return all key values which match those metadata, so that > > > >> operator > > > >> > can mange key in their own understading to a distributed system in > > > >> > seperated views. > > > >> > *config batch action: * with metadate, you can batch update, > delete > > > key > > > >> > values. for example: update all "service=cart version=1.0" > > > configuration > > > >> > log level to "DEBUG" or just update "service=cart version=1.0, > > > >> ip=x.x.x.x" > > > >> > . > > > >> > *rich value type: *not ony plain text any more, but support to be > > > aware > > > >> of > > > >> > ini, json,yaml,xml and java properties > > > >> > *heterogenous configuration system: *able to manage configuration > in > > > k8s > > > >> > and consul or even more, you can update, delete, and use config > view > > > for > > > >> > those systems, and you can integrate with your own config system > to > > > >> > MetaConfig by following standardized API and model > > > >> > > > > > > >
