nginx kerberos module

turns out with this module (comes with nginx-plus with less installation hassle), the (almost) only required config is to provide the keytab

auth_gss on;
auth_gss_allow_basic_fallback off;
auth_gss_keytab <path_to_keytab>;

jupyter shell vs kernel

if the shell (where the notebook is started) is different from the kernel executable, when using

pip install
conda install

it could install to a directory where the python executable in the notebook is not referring to.

the solution is

# check the python shell
!type python

# check the executable path

when install, make sure it align, or using the command

!conda install --yes --prefix {sys.prefix} numpy

!{sys.executable} -m pip install numpy

sql nolock

nolock is a table hint. when running with a select query, its the same as (



however, seems like besides resulting dirty reads, it could also cause an issue with mssql driver 10+ (at the moment) to have to `read query timeout`.

i am still trying to dig out the reason for the timeout, but nolock or readuncommited is definitely not suggested.

junit 5 with mockito for partial test

I have a class which has a number of attributes initiated at construction (this arguable could be a code smell). There is a need to test one function (methodA) of the class, which leverages several its attributes.

Without much interest in each attributes functioning, those attributes could be mocked.

However, there would a lot methods to set up individually for the attributes and its functions used in methodA, which basically raised a need to test a function partially.

Here is the way I made it work:

mock for the class and function to test :

OrderService order; //the interested class, to be tested

UserService user; //attribute of OrderService

PaymentService payment; //attribute of OrderService

set up the fields (with reflection) and stop further processing (halfway of methodA)

//the core part is for `methodA` to invoke the realMethod

//then to stop it halfway
when(payment.process()).then(...NPE with corresponding message);

finally to verify the test result with assertJ

handle null from json with jackson

when jackson deserialize the json into object, some of the optional fields (non `@Notnull`) could get a null value assigned, something like

 "field1": null

which would result into

class Payload{

  String field1;

where when the payload object being used, it could throw out NPE if the field1 is not handled properly.

The way to handle this properly is through setting a default value during deserialization, something like

class Payload{

  String field1="DEFAULT-NULL";