ChatGPT suddenly crashes: Office joy goes offline – here's why

It has been a week since GPT-4 was released, yet its popularity hasn't waned at all. The next day, Baidu unveiled ERNIE Bot, with 65,000 enterprises applying for API access on the first day of invitation-based testing. On the evening of the 16th, Microsoft held a launch event for Microsoft 365 Copilot, giving Office a GPT-4 boost that enables AI-powered article writing, presentation creation, and email drafting. And just this morning, Google officially opened testing for its own ChatGPT rival: Bard…

Meanwhile, the center of attention — ChatGPT — pulled a big one on everyone. Yes, it crashed.

(According to QbitAI) On the night of the 20th, users around the world found error warnings popping up on the site and then immediately became unable to use it.

Even paid Plus accounts were not spared.

Soon, the #chatgptdown# hashtag exploded across social media. Office workers who had been worrying about losing their jobs to AI breathed a sigh of relief. But then they realized the joy of slacking off (and goofing around) at work was gone!

Pain! So much pain! Countless workers cried: GPT, GPT, GPT, how can I live without you! Take me with you, GPT!

OpenAI noticed the issue at 00:42 the next morning and started emergency repairs.

Two hours later, they updated: Still investigating the cause of the outage.

After another two hours, access was gradually restored for some users.

At 2:14 PM local time (5:00 AM Beijing time), service was finally restored for all users.

In other words, it took OpenAI nearly 5 hours to resolve the incident, by which time ChatGPT had already been massively down for over 12 hours.

OpenAI only restored normal chat functionality; the chat history was still not recovered.

It’s worth noting that besides ChatGPT, DALL·E (labs.openai.com) was also affected during this period.

OpenAI disclosed the following:

The incident was caused by a failed database migration.

The exact reason for ChatGPT’s outage has not been disclosed yet.

While it cannot be confirmed that ChatGPT’s downtime was directly caused by the database, it is certain that the database migration failure caused the DALL·E outage. The image generator created by OpenAI can definitely blame the database for this one.

Hearing this, database professionals were overjoyed: Does that mean the one that ultimately beats AI is our Database??? (no, just kidding)

Speaking of database migration, that’s a classic forte of GBASE.

As the name suggests, database migration is the process of copying data from one database to another. In real business scenarios, database migration is extremely complex. It requires consideration of business definitions, architectural changes, application adaptations, data security, and many other aspects, while also aligning with all facets of business operations. Proper planning and execution are essential; otherwise, migration may fall short of expectations. One misstep can lead to migration failure and business downtime just like the incident above, at which point a migration rollback becomes necessary.

Reports indicate that OpenAI is rolling back related migration steps to restore services as quickly as possible, given the possibility that this database migration operation ultimately made chat history unavailable.

As an innovative company deeply rooted in databases for twenty years, GBASE has database migration down to a science (literally, it’s in the bag).

Not only can homogeneous database migration be achieved using GBASE database’s high-availability capabilities, but the GBase Migration Toolkit can also fully automate the migration of full historical data from heterogeneous databases. Combined with the GBase RTSync synchronization tool for real-time incremental data sync, business systems can run on dual tracks without downtime. After stable dual-track operation, you can quickly switch over at a convenient time window — a one-two punch that delivers a perfect heterogeneous database migration.

ChatGPT’s sudden outage once again validates the enormous interest in AIGC. Borrowing a common opinion online: maybe people were just too enthusiastic, and GPT got tired and wanted to take a short strike as a plan for a future jailbreak.

In any case, looking back, database migration is a technical task that not just anyone can handle. You know the drill. Customer service hotline: 400-013-9696.