mirror of
https://github.com/Foundry376/Mailspring.git
synced 2024-11-11 18:32:20 +08:00
4274753271
This should prevent us from flooding the app with contacts (/contacts responds very quickly). Fixes T3475
248 lines
9.5 KiB
CoffeeScript
248 lines
9.5 KiB
CoffeeScript
_ = require 'underscore'
|
|
{Actions, DatabaseStore, Account, Thread} = require 'nylas-exports'
|
|
NylasLongConnection = require '../lib/nylas-long-connection'
|
|
NylasSyncWorker = require '../lib/nylas-sync-worker'
|
|
|
|
describe "NylasSyncWorker", ->
|
|
beforeEach ->
|
|
@apiRequests = []
|
|
@api =
|
|
accessTokenForAccountId: =>
|
|
'123'
|
|
makeRequest: (requestOptions) =>
|
|
@apiRequests.push({requestOptions})
|
|
getCollection: (account, model, params, requestOptions) =>
|
|
@apiRequests.push({account, model, params, requestOptions})
|
|
getThreads: (account, params, requestOptions) =>
|
|
@apiRequests.push({account, model:'threads', params, requestOptions})
|
|
|
|
spyOn(DatabaseStore, 'persistJSONObject').andReturn(Promise.resolve())
|
|
spyOn(DatabaseStore, 'findJSONObject').andCallFake (key) =>
|
|
expected = "NylasSyncWorker:#{TEST_ACCOUNT_ID}"
|
|
return throw new Error("Not stubbed! #{key}") unless key is expected
|
|
Promise.resolve _.extend {}, {
|
|
"contacts":
|
|
busy: true
|
|
complete: false
|
|
"calendars":
|
|
busy:false
|
|
complete: true
|
|
}
|
|
|
|
@account = new Account(clientId: TEST_ACCOUNT_CLIENT_ID, serverId: TEST_ACCOUNT_ID, organizationUnit: 'label')
|
|
@worker = new NylasSyncWorker(@api, @account)
|
|
@connection = @worker.connection()
|
|
advanceClock()
|
|
|
|
it "should reset `busy` to false when reading state from disk", ->
|
|
@worker = new NylasSyncWorker(@api, @account)
|
|
spyOn(@worker, 'resumeFetches')
|
|
advanceClock()
|
|
expect(@worker.state().contacts.busy).toEqual(false)
|
|
|
|
describe "start", ->
|
|
it "should open the long polling connection", ->
|
|
spyOn(@connection, 'start')
|
|
@worker.start()
|
|
advanceClock()
|
|
expect(@connection.start).toHaveBeenCalled()
|
|
|
|
it "should start querying for model collections and counts that haven't been fully cached", ->
|
|
spyOn(@connection, 'start')
|
|
@worker.start()
|
|
advanceClock()
|
|
expect(@apiRequests.length).toBe(8)
|
|
modelsRequested = _.compact _.map @apiRequests, ({model}) -> model
|
|
expect(modelsRequested).toEqual(['threads', 'contacts', 'drafts', 'labels'])
|
|
|
|
countsRequested = _.compact _.map @apiRequests, ({requestOptions}) ->
|
|
if requestOptions.qs?.view is 'count'
|
|
return requestOptions.path
|
|
|
|
expect(modelsRequested).toEqual(['threads', 'contacts', 'drafts', 'labels'])
|
|
expect(countsRequested).toEqual(['/threads', '/contacts', '/drafts', '/labels'])
|
|
|
|
it "should mark incomplete collections as `busy`", ->
|
|
@worker.start()
|
|
advanceClock()
|
|
nextState = @worker.state()
|
|
|
|
for collection in ['contacts','threads','drafts', 'labels']
|
|
expect(nextState[collection].busy).toEqual(true)
|
|
|
|
it "should initialize count and fetched to 0", ->
|
|
@worker.start()
|
|
advanceClock()
|
|
nextState = @worker.state()
|
|
|
|
for collection in ['contacts','threads','drafts', 'labels']
|
|
expect(nextState[collection].fetched).toEqual(0)
|
|
expect(nextState[collection].count).toEqual(0)
|
|
|
|
it "after failures, it should attempt to resume periodically but back off as failures continue", ->
|
|
simulateNetworkFailure = =>
|
|
@apiRequests[1].requestOptions.error({statusCode: 400})
|
|
@apiRequests = []
|
|
|
|
spyOn(@worker, 'resumeFetches').andCallThrough()
|
|
@worker.start()
|
|
|
|
expect(@worker.resumeFetches.callCount).toBe(1)
|
|
simulateNetworkFailure(); expect(@worker.resumeFetches.callCount).toBe(1)
|
|
advanceClock(30000); expect(@worker.resumeFetches.callCount).toBe(2)
|
|
simulateNetworkFailure(); expect(@worker.resumeFetches.callCount).toBe(2)
|
|
advanceClock(30000); expect(@worker.resumeFetches.callCount).toBe(2)
|
|
advanceClock(30000); expect(@worker.resumeFetches.callCount).toBe(3)
|
|
simulateNetworkFailure(); expect(@worker.resumeFetches.callCount).toBe(3)
|
|
advanceClock(30000); expect(@worker.resumeFetches.callCount).toBe(3)
|
|
advanceClock(30000); expect(@worker.resumeFetches.callCount).toBe(4)
|
|
simulateNetworkFailure(); expect(@worker.resumeFetches.callCount).toBe(4)
|
|
advanceClock(30000); expect(@worker.resumeFetches.callCount).toBe(4)
|
|
advanceClock(30000); expect(@worker.resumeFetches.callCount).toBe(4)
|
|
advanceClock(30000); expect(@worker.resumeFetches.callCount).toBe(5)
|
|
|
|
describe "when a count request completes", ->
|
|
beforeEach ->
|
|
@worker.start()
|
|
advanceClock()
|
|
@request = @apiRequests[0]
|
|
@apiRequests = []
|
|
|
|
it "should update the count on the collection", ->
|
|
@request.requestOptions.success({count: 1001})
|
|
nextState = @worker.state()
|
|
expect(nextState.threads.count).toEqual(1001)
|
|
|
|
describe "resumeFetches", ->
|
|
it "should fetch collections", ->
|
|
spyOn(@worker, 'fetchCollection')
|
|
@worker.resumeFetches()
|
|
expect(@worker.fetchCollection.calls.map (call) -> call.args[0]).toEqual(['threads', 'calendars', 'contacts', 'drafts', 'labels'])
|
|
|
|
describe "fetchCollection", ->
|
|
beforeEach ->
|
|
@apiRequests = []
|
|
|
|
it "should not start if the collection sync is already in progress", ->
|
|
@worker._state.threads = {
|
|
'busy': true
|
|
'complete': false
|
|
}
|
|
@worker.fetchCollection('threads')
|
|
expect(@apiRequests.length).toBe(0)
|
|
|
|
it "should not start if the collection sync is already complete", ->
|
|
@worker._state.threads = {
|
|
'busy': false
|
|
'complete': true
|
|
}
|
|
@worker.fetchCollection('threads')
|
|
expect(@apiRequests.length).toBe(0)
|
|
|
|
it "should start the request for the model count", ->
|
|
@worker._state.threads = {
|
|
'busy': false
|
|
'complete': false
|
|
}
|
|
@worker.fetchCollection('threads')
|
|
expect(@apiRequests[0].requestOptions.path).toBe('/threads')
|
|
expect(@apiRequests[0].requestOptions.qs.view).toBe('count')
|
|
|
|
it "should start the first request for models", ->
|
|
@worker._state.threads = {
|
|
'busy': false
|
|
'complete': false
|
|
}
|
|
@worker.fetchCollection('threads')
|
|
expect(@apiRequests[1].model).toBe('threads')
|
|
expect(@apiRequests[1].params.offset).toBe(0)
|
|
|
|
describe "when an API request completes", ->
|
|
beforeEach ->
|
|
@worker.start()
|
|
advanceClock()
|
|
@request = @apiRequests[1]
|
|
@apiRequests = []
|
|
|
|
describe "successfully, with models", ->
|
|
it "should request the next page", ->
|
|
pageSize = @request.params.limit
|
|
models = []
|
|
models.push(new Thread) for i in [0..(pageSize-1)]
|
|
@request.requestOptions.success(models)
|
|
advanceClock(2000)
|
|
expect(@apiRequests.length).toBe(1)
|
|
expect(@apiRequests[0].params).toEqual
|
|
limit: pageSize,
|
|
offset: @request.params.offset + pageSize
|
|
|
|
it "should update the fetched count on the collection", ->
|
|
expect(@worker.state().threads.fetched).toEqual(0)
|
|
pageSize = @request.params.limit
|
|
models = []
|
|
models.push(new Thread) for i in [0..(pageSize-1)]
|
|
@request.requestOptions.success(models)
|
|
expect(@worker.state().threads.fetched).toEqual(pageSize)
|
|
|
|
describe "successfully, with fewer models than requested", ->
|
|
beforeEach ->
|
|
models = []
|
|
models.push(new Thread) for i in [0..100]
|
|
@request.requestOptions.success(models)
|
|
|
|
it "should not request another page", ->
|
|
expect(@apiRequests.length).toBe(0)
|
|
|
|
it "should update the state to complete", ->
|
|
expect(@worker.state().threads.busy).toEqual(false)
|
|
expect(@worker.state().threads.complete).toEqual(true)
|
|
|
|
it "should update the fetched count on the collection", ->
|
|
expect(@worker.state().threads.fetched).toEqual(101)
|
|
|
|
describe "successfully, with no models", ->
|
|
it "should not request another page", ->
|
|
@request.requestOptions.success([])
|
|
expect(@apiRequests.length).toBe(0)
|
|
|
|
it "should update the state to complete", ->
|
|
@request.requestOptions.success([])
|
|
expect(@worker.state().threads.busy).toEqual(false)
|
|
expect(@worker.state().threads.complete).toEqual(true)
|
|
|
|
describe "with an error", ->
|
|
it "should log the error to the state", ->
|
|
err = new Error("Oh no a network error")
|
|
@request.requestOptions.error(err)
|
|
expect(@worker.state().threads.busy).toEqual(false)
|
|
expect(@worker.state().threads.complete).toEqual(false)
|
|
expect(@worker.state().threads.error).toEqual(err.toString())
|
|
|
|
it "should not request another page", ->
|
|
@request.requestOptions.error(new Error("Oh no a network error"))
|
|
expect(@apiRequests.length).toBe(0)
|
|
|
|
it "resumes when a action forces it to", ->
|
|
err = new Error("Oh no a network error")
|
|
@request.requestOptions.error(err)
|
|
expect(@worker.state().threads.busy).toEqual(false)
|
|
expect(@worker.state().threads.complete).toEqual(false)
|
|
spyOn(@worker, 'resumeFetches').andCallThrough()
|
|
Actions.retryInitialSync()
|
|
expect(@worker.resumeFetches).toHaveBeenCalled()
|
|
expect(@worker.resumeFetches.calls.length).toBe 1
|
|
expect(@worker.state().threads.busy).toEqual(true)
|
|
expect(@worker.state().threads.error).toBe(null)
|
|
|
|
describe "cleanup", ->
|
|
it "should termiate the long polling connection", ->
|
|
spyOn(@connection, 'end')
|
|
@worker.cleanup()
|
|
expect(@connection.end).toHaveBeenCalled()
|
|
|
|
it "should stop trying to restart failed collection syncs", ->
|
|
spyOn(console, 'log')
|
|
spyOn(@worker, 'resumeFetches').andCallThrough()
|
|
@worker.cleanup()
|
|
advanceClock(50000)
|
|
expect(@worker.resumeFetches.callCount).toBe(0)
|