跟讀練習: Interview with Big Data engineer in 2026. - 通過影片學習英語口語
選擇練習模式
每種模式都會開啟全屏練習室。
168 句
如果句子過短或過長，請點選 Edit 進行調整。
1
We built a real -time pipeline
2
So leadership can ignore insights at millisecond speeds
3
What was the question?
4
Okay, sorry.
5
I had quick
6
Problem in database, but no worries all 200 tetabytes are recovered
7
Big data most companies don't actually have big data
8
They have medium data.
9
Big data is anything that crushes Excel.
10
We actually have two big data for fault tolerance.
11
Two wrongs don't make a right, except in a distributed database.
12
Big data is different math.
13
1 plus 1, 1 .9.
14
95 % confidence interval.
15
Our 15 million raw data set has 800 columns.
16
The rows aren't wide, they are panoramic.
17
If the joint fits in memory, it belongs in Postgres.
18
Our architecture preserves every invariant except the budget.
19
Our cloud bill was $340 ,000.
20
We don't know if the processing even ran.
21
That was just observability.
22
We support real -time processing, of course.
23
Unless the event arrives early, late, duplicated, malformed, reordered, eventual consistency was too hard.
24
So we went with immediately inaccurate.
25
70 % of records are accurate.
26
It's higher than the weather forecast.
27
The numbers are not wrong.
28
They are early.
29
Wait, two, three, maybe four days, they will be late.
30
It's almost interactive analytics.
31
Final numbers mean numbers we have stopped refreshing.
32
The slide says we handle billions of events.
33
It does not say we handle them well.
34
It is only truly idempotent if all availability zones go down equally during the outage.
35
Management.
36
Management asked us to build a data lake, but it wasn't scalable enough.
37
So McKinsey asked us to build a data swamp.
38
The project was declared successful before all the metrics arrived.
39
We call that watermarking.
40
Dashboard is green because the job that checks if the dashboard is wrong is red.
41
Ah, what happened here?
42
Let me quickly roll back 30 billion events.
43
A production workaround becomes durable the moment the engineer who wrote it leaves.
44
Yes.
45
We removed an unused column and it broke Portugal.
46
The bug has existed for eight years.
47
At this point it's an interface.
48
Quick Slack message.
49
20 terabytes lost again.
50
This is our architecture.
51
Very simple.
52
Kafka, Airflow, Databricks, Flink, Spark, SQL, Iceberg, S3, Presto, DBT.
53
What does DBT stands for dialect behavioral therapy.
54
We need that to avoid going crazy.
55
We have the most modern stack in the industry.
56
Only 10 % is COBOL.
57
HDFS is not dead.
58
It's waiting for someone with root access.
59
We migrated from Kroon to Airflow for reliability.
60
Then we wrote a Kroon job to check if Airflow silently died.
61
We would switch from Databricks to Excel if it wasn't for date parsing.
62
We are a data driven company of course.
63
Every week somebody reads the invoices out of PDFs and types the data into Excel.
64
Running on Kubernetes.
65
We moved from Hive tables to Iceberg tables because we wanted the metadata to fail with better governance.
66
Last week, we had some ransomware attack.
67
We lost the backups, yes.
68
Whenever we lose a backup, we reconstruct what the customer did from the logs.
69
No, that's one of our industry's best practices.
70
Our customers didn't even know.
71
To be honest, most of our customers don't even use our product anyway.
72
The board deck says petabyte scale, yeah.
73
Scale is a direction, not a quantity.
74
I checked with a legal.
75
Management says one petabyte.
76
I checked, yeah.
77
We only process 18 terabytes, but we round up.
78
Our P99 is four minutes.
79
But the P50 is excellent.
80
And most people are median.
81
Streaming gets simple once you understand watermarks.
82
Event time, processing time, state, back pressure.
83
Batch and streaming produce different numbers.
84
We average them.
85
Our recommendation pipeline had three hours of Kafka lag.
86
So technically it was a real time communication for your past self.
87
Kafka rebalanced so often I started listing it under team building activities.
88
We decoupled storage and compute.
89
Now the storage team and the compute team have separate outages.
90
Ah, again, we lost the logs from Tuesday.
91
The data team, the data team has excellent isolation.
92
We haven't spoken to product in seven months.
93
All problems become your problems at one terabyte an hour.
94
Exactly once processing is easy.
95
You just need two topics, transactions, offsets, recovery state and a relaxed interpretation of easy.
96
GDPR in a data warehouse.
97
This is a funny joke.
98
We anonymize the data by removing the column name.
99
We persist it in places nobody knows how to query.
100
The only thing in this company with strong durability is a bad architectural decision.
101
Salary, job, security, holidays, pick two, a CIP compliant.
102
Can we have this by Friday?
103
In distributed systems, what does Friday mean?
104
Before the meeting is only a partial order.
105
The project isn't late.
106
The clocks just disagree.
107
It's a distributed organization.
108
Every quarter we solve capacity problems by increasing input.
109
We maintain system availability by shedding JIRA tickets.
110
Last year, they fired 80 % of our team as part of our operational transformation.
111
They should have used CRDTs instead.
112
A backfill is when you fix one historical bug by introducing a brand new historical bug.
113
The Snowflake warehouse runs a two -second query every minute.
114
A very sophisticated way of never letting it turn off.
115
Artificial intelligence?
116
No. We still rely on humans, a single point of failure.
117
Yes, David.
118
The game.
119
You pull something from the bottom, everything falls apart.
120
It's educational game for industry.
121
Back 2016, learning Hadoop in the simple 760 steps.
122
That's right.
123
Hadoop certified.
124
Cloudera certified.
125
Cloudera no longer offers this cert.
126
We migrated to Abachi iceberg this year.
127
Quick 43 months migration.
128
We still use protobuf.
129
Yes, we are still confused, but at least confused strongly type.
130
Some say data is the new oil.
131
We cannot stop producing it and we will pay to clean it up for years.
132
From what I see, data is the new asbestos.
133
The AI team spent four million and it called investment.
134
I spent three hundred forty thousand.
135
It is called a review.
136
I'm in FinOps review every Thursday.
137
The AI team has never ever been to FinOps.
138
They have a different word for money.
139
My cluster is a cost center.
140
Their cluster is a strategy.
141
Same rack in AWS where I make, there are my glasses.
142
Duh!
143
Ah, only disk usage 80%.
144
This is sweet spot.
145
A soft outage is when the checkpoint directory grows faster than S3 versioning can build.
146
Back in the days, big data means you bring the intelligence to the data.
147
That's why I go straight to big data center.
148
We have two big data centers for fault tolerance.
149
I wanted to become simple app developer.
150
Take app, build it, host it on railway, no problems.
151
No thinking required.
152
I can go on holidays.
153
But the team came with Airflow Graph.
154
We use what?
155
1, 2, 3, 4, 13, 14, 23, 60, 400 services just to clean the locks.
156
Whole dashboard?
157
You need whole company.
158
I am on call today, yes.
159
As you can see.
160
Oh, I turn my phone off.
161
If it burns, it burns.
162
I don't have keys to service anyway.
163
What am I gonna do?
164
I'm not ignoring Slack.
165
I'm just following our load shedding policy.
166
Don't ship late and don't ship too early.
167
Ship peacefully.
168
Thank you, Railway, for sponsoring this video.
同頻道
FOR YOU
推薦影片
C11
358 句
It’s not just you
Daily ConversationUniversity StudentsWorking Adults
C11
77 句
The Secret Sauce Behind NoSQL: LSM Tree
Daily ConversationUniversity StudentsWorking Adults
C13
134 句
How Uber Scaled from One Server to 150 Million Reads Per Second
Daily ConversationUniversity Students
C12
118 句
The Anatomy of Fraud
Daily ConversationUniversity StudentsWorking Adults
C11
398 句
Sec Growth DataScience staff meeting Sep 14 2022
Daily ConversationUniversity StudentsWorking Adults
C11
689 句
Database Sharding! Designing Data-Intensive Applications chapter 6
Daily ConversationUniversity StudentsWorking Adults
C138 句
"Dive Deep" Leadership Principle Explained by Amazon CEO Andy Jassy
Daily ConversationUniversity StudentsWorking Adults
C11
389 句
They JUST Accidentally Revealed the Truth: The AI Bubble
Daily ConversationUniversity StudentsWorking Adults
B21
220 句
How Modern Data Teams Work (Engineer, Analyst, Scientist, Architect, ML)
Daily ConversationUniversity Students
C1231 句
Redis Crash Course - the What, Why and How to use Redis as your primary database
Daily ConversationUniversity StudentsWorking Adults
C11
532 句
How an Ex-Googler Accidentally Founded a Database Company
Daily ConversationUniversity StudentsWorking Adults
C11
37 句
How Anthropic uses Claude in Product Management
Daily ConversationUniversity StudentsWorking Adults
本課的詞彙與口語要點
這節口語課以影片“Interview with Big Data engineer in 2026”為素材。 這段影片共有 168 個句子、1183 個單詞可供跟讀。 講話部分時長為 8:34。 說話人語速平穩，每分鐘約 138 個詞，很適合跟讀。 只有 76% 的單詞屬於英語最常用的 3,000 詞，詞彙難度較高。
影片中的重點詞彙
影片中 15 個值得學習的單詞，附發音和釋義：
| 單詞 | 發音 | 釋義 |
|---|---|---|
| distribute 動詞 | /dɪˈstɹɪbjuːt/ | 分配 |
| ignore 動詞 | /ɪɡˈnɔː/ | 無視 /無視, 忽略 |
| confused 形容詞 | /kənˈfjuːzd/ | 困惑 |
| railway 名詞 | /ˈɹeɪlˌweɪ/ | 鐵路 /鐵路, 鐵道 /鐵道 |
| architecture 名詞 | /ˈɑː.kɪˌtɛk.t͡ʃə/ | 建築 /建築, 建築學 /建築學 |
| log 名詞 | /lɔɡ/ | 圓木 /圓木 |
| column 名詞 | /ˈkɒləm/ | 柱子, 柱 |
| database 名詞 | /ˈdeɪtəˌbeɪs/ | 資料庫 /資料庫 |
| backup 名詞 | /ˈbækˌʌp/ | 備份 /備份 |
| availability 名詞 | /əˌveɪ.ləˈbɪl.ɪ.ti/ | 可及性 |
| warehouse 名詞 | /ˈwɛə(ɹ)haʊs/ | 倉庫 /倉庫 |
| tolerance 名詞 | /ˈtɑ.lə.ɹəns/ | 寬容 /寬容 |
| pipeline 名詞 | /ˈpaɪpˌlaɪn/ | 管道, 管線 /管線 |
| cluster 名詞 | /ˈklʌstɚ/ | 串, 叢 /叢 |
| query 名詞 | /ˈkwɪɹ.i/ | 問題 /問題, 質問 /質問 |
影片中出現的短語動詞
| 單詞 | 釋義 |
|---|---|
| go down 動詞 | 下降, 下去 |
| turn off 動詞 | 關 /關, 關掉 /關掉 |
值得反覆跟讀的句子
影片中簡短而完整的句子，可以直接用在日常對話裡：
- We don't know if the processing even ran.
- No, that's one of our industry's best practices.
- Our customers didn't even know.
- We haven't spoken to product in seven months.
- What am I gonna do?
影片中的語法
說話人最常用的結構，並附上影片中的原話：
| 結構 | 影片中的用法 |
|---|---|
| 現在完成時 have/has + 過去分詞 — 過去發生但與現在仍有關聯的事 | have stopped · has existed · haven't spoken |
| 被動語態 be + 過去分詞 — 強調發生了什麼，而不是誰做的 | are recovered · was declared · is called |
需要注意的發音
說話人用了 17 次縮略和弱讀形式，例如 don't, I'm, wasn't。請按聽到的簡短形式來說。
- “sh” 和 “zh” 音: artificial /ˌɑː.tɪˈfɪʃ.əl/, interpretation /ɪnˌtɜːpɹəˈteɪʃən/, operational /ˌɑ.pəˈɹeɪ.ʃə.nəl/, transaction /tɹænˈzækʃən/, transformation /ˌtɹæns.fɔɹˈmeɪ.ʃən/
- 長單詞——注意重音位置: architecture /ˈɑː.kɪˌtɛk.t͡ʃə/, availability /əˌveɪ.ləˈbɪl.ɪ.ti/, artificial /ˌɑː.tɪˈfɪʃ.əl/, interpretation /ɪnˌtɜːpɹəˈteɪʃən/, operational /ˌɑ.pəˈɹeɪ.ʃə.nəl/
中文母語者容易讀錯的音：
- /v/ — 不要讀成 /w/，上齒輕觸下唇: availability /əˌveɪ.ləˈbɪl.ɪ.ti/, solve /sɒlv/, arrive /əˈɹaɪv/, recover /ɹɪˈkʌvɚ/, developer /dɪˈvɛləpɚ/
- 詞尾子音 — 要讀清楚，後面不要加母音: distribute /dɪˈstɹɪbjuːt/, confused /kənˈfjuːzd/, log /lɔɡ/, database /ˈdeɪtəˌbeɪs/, backup /ˈbækˌʌp/
- /r/ 和 /l/ 的區別: railway /ˈɹeɪlˌweɪ/, tolerance /ˈtɑ.lə.ɹəns/, strongly /ˈstɹɔŋli/, salary /ˈsæləɹi/, relax /ɹɪˈlæks/
如何用這段影片練習
- 先完整聽一遍影片，不要開口，記下不認識的單詞。
- 用正常速度逐句跟讀，每句重複到你的節奏與說話人一致為止。
- 錄下自己的聲音並與原聲對比，特別注意 distribute, ignore, confused 這類單詞。
什麼是跟讀法？
跟讀法 (Shadowing) 是一種有科學依據的語言學習技巧，最初開發用於專業口譯員的培訓，並由多語言者Alexander Arguelles博士普及。這個方法簡單而強大：您在聽英語母語原聲的同時立即大聲重複——就像是一個延遲1-2秒緊跟說話者的影子。與被動聽力或語法練習不同，跟讀法強迫您的大腦和口腔肌肉同時處理並模模擬實的講話模式。研究表明它能顯著提高發音準確性，語調，節奏，連讀，聽力理解和口語流利度——使其成為雅思口語備考和真實英語交流最有效的方法之一。