This issue threw me for a loop for some time.

Suppose you have a Schedulable class and when it runs, the behavior of the execute() method doesn’t match what you expect. It looks like there’s memory of previous execute()s

Here is a specific example that distills the issue

public class MySchedulable implements Schedulable {

   Account[] accts = new List<Account>();

   public void execute(SchedulableContext sc) {
      accts.addAll([select id, name from Account where YTD_Total__c < 1000.0]);
      // do something interesting with this 
   }

You run the job weekly. You observe on week n, where n > 1, that Accounts are being processed that currently don’t have a YTD_Total__c < 1000. Accounts with YTD_Total__c > 1000 are being processed (?!?)

Explanation

• When the job is scheduled, the class constructor is called (here, the implicit constructor) and the object variable accts is initialized empty.
• On week 1, when execute() is called, the accounts with YTD_Total__c < 1000 as of week 1 are added to list accts.
• On week 2, when execute() is called, the accounts with YTD_Total__c < 1000 as of week 2 are added to list accts.

Note that the class constructor is not reinvoked for each execute(). Hence, the accts list grows and grows.

Solution
Reinitialize all object variables within the execute().

public class MySchedulable implements Schedulable {
 
   public void execute(SchedulableContext sc) {
      Account[] accts = new List<Account>();
      accts.addAll([select id, name from Account where YTD_Total__c < 1000.0]);
      // do something interesting with this 
   }