This guide is for users who already have their own server details and want to use Hysteria2 in Shadowrocket. First, check how to get the app for your device. Then distinguish the roles of UDP, QUIC, and congestion control. Finally, import the connection, isolate routing rules, and compare results under the same conditions to see whether the protocol suits your network. A successful latency test does not prove that the Hysteria2 handshake works, and high packet loss does not automatically mean switching protocols will increase throughput.
1. Check device availability, app access, and server requirements
The iPhone and iPad version of Shadowrocket is a one-time $2.99 purchase on the App Store. The developer is Shadow Launch Technology Limited; check the App Store listing for system requirements. The Mac (Apple silicon and Intel), Windows, Android, and Linux versions are free. Visit the download page for all platforms to choose the right option for your device, then use your existing server details.
Installing the client and obtaining a server you can connect to are separate things. A one-time app purchase is not a service plan. This guide does not provide, sell, or recommend servers or subscriptions. Hysteria2 requires a compatible protocol on both the server and client; you cannot simply change the Type of a Shadowsocks, VMess, VLESS, Trojan, or WireGuard entry to Hysteria2 and expect it to work.
- Server support: The remote server must actually run a compatible Hysteria2 service and provide the correct server address, UDP port, and authentication details.
- Certificate identity: Obtain the certificate domain and any required SNI details from the service provider. The connection IP and certificate verification name may differ.
- Additional parameters: If obfuscation is enabled on the server, you need the matching obfuscation method and password. The authentication password and obfuscation password are not interchangeable.
- Network requirements: Your local network, router, and server firewall must all allow the required UDP traffic. A reachable TCP port does not mean the corresponding UDP port is reachable.
2. What UDP, QUIC, and Hysteria2 each do
UDP provides datagram transport but does not itself handle reliable retransmission, in-order delivery, or congestion control. QUIC runs over UDP and implements encrypted connections, acknowledgments, loss recovery, and multiplexed streams at the transport layer. Hysteria2 uses QUIC for proxy transport and adds authentication, proxy requests, and its own traffic-control design. So using UDP does not mean data is simply discarded, nor does it mean applications no longer need reliable transport.
For proxied TCP traffic, Hysteria2 carries reliable byte streams over QUIC streams. If a stream loses data, that stream may still wait for recovery. The benefit is that other streams do not all have to wait for a gap in the same underlying TCP byte stream. However, the streams share the connection's congestion control and available bandwidth; they are not completely independent dedicated links.
For UDP traffic, Hysteria2 uses datagrams to carry the relevant data; it does not guarantee that every application-layer datagram becomes a reliable message. Voice and other real-time applications are still affected by jitter, queueing, and their own recovery mechanisms. A QUIC connection handshake also does not complete every handshake needed to reach a destination website: after the proxy is established, the website's DNS, TCP, or TLS stages may still take time.
3. Understand congestion control to see why it can be fast
Throughput is not determined by a protocol's name alone. Traditional loss-based congestion control may treat random packet loss as a sign of congestion and reduce the sending rate. On some long-distance or wireless paths with significant interference, that response can limit throughput. Hysteria2's Brutal congestion control works toward a target bandwidth and adjusts sending based on loss feedback, aiming to maintain target goodput in the presence of random packet loss.
This approach does not create extra network capacity. If the bottleneck is an already saturated link, sending more traffic to compensate can increase queueing, jitter, and wait times for other connections. A higher target bandwidth is not necessarily better, and the advertised speed of your broadband plan should not be treated as the sustained rate available to a remote server.
- Bandwidth negotiation has prerequisites: Hysteria2's congestion-control mode depends on client bandwidth information, server limits, and negotiation policy. The Type setting alone does not prove that the current connection is using Brutal.
- Other control modes are available: The protocol also supports operation based on BBR. Whether it is used, and whether the server accepts a bandwidth value proposed by the client, depends on both implementations and the server configuration.
- Keep directions straight: Download and upload may have different bottlenecks. Mbps and MB/s are also different units and should not be entered as though the numbers were equivalent.
- The interface is not the entire protocol: Options in server documentation may not appear individually in Shadowrocket's current editing interface. Do not enter configuration keys from other programs into unrelated fields.
Bottom line: identify the bottleneck before adjusting bandwidth
If raising the sending target only increases latency and packet loss without improving goodput, lower the target and test again instead of raising it further. Also check the server's bandwidth limit and the congestion-control mode actually in use.
4. Networks worth testing with Hysteria2
To assess whether it fits, first distinguish random packet loss, persistent congestion, and UDP restrictions. All three can make websites feel slow, but they call for different fixes. High latency only means that round trips take longer; by itself, it does not show that Hysteria2 will be faster. High packet loss also needs to be assessed in context.
| Network symptom | Possible effect | What to check |
|---|---|---|
| Intermittent packet loss on Wi-Fi | Reliable transport may trigger recovery, causing throughput to fluctuate | Test from a fixed location; compare packet loss and jitter when idle and during transfers |
| Long-distance path, high RTT | Longer handshake and feedback cycles | Record connection setup time and sustained download throughput separately |
| Persistent congestion at the network exit | Compensating sends may increase queueing | Lower the sending target and see whether interactive latency drops |
| UDP restricted or dropped | QUIC may fail to handshake or stall mid-connection | Check the destination UDP port, then test over Wi-Fi and cellular |
| Low packet loss, stable bandwidth | Differences between protocols may be minimal | Decide whether to keep it based on connection stability and time to complete real tasks |
One telltale pattern: the same server details connect over cellular but keep timing out on a particular Wi-Fi network. First check how that network handles UDP rather than changing the password. Another pattern: transfer speeds start high, then web interactions become noticeably slower. That points to queueing at the bottleneck and the target bandwidth as areas to investigate—not just the peak speed.
5. Import Hysteria2 into Shadowrocket and isolate routing variables
The steps below use the interface labels Home, Config, and Settings. Layouts may vary by platform; the fields you can edit depend on the actual interface and the server details you have. Set up a complete connection first, then address routing rules so you do not change multiple variables at once.
-
Import your service
In Home, tap “+” in the upper-right corner → Type → Subscribe to import a subscription you already have. After saving, update it from Home. To add a connection manually, choose Hysteria2 under Type and enter the details supplied by your service provider. Do not reuse ports or authentication fields from another protocol.
-
Check the parameters
In Home, open the entry's edit screen and check the address, port, authentication, certificate domain, and obfuscation details one by one. If the server uses a domain, make sure that domain resolves. Do not enter a subscription URL as the server address.
-
Isolate routing rules
In Home → Global Routing, temporarily select Proxy and make sure the test request actually uses the Hysteria2 entry you are testing. If the request goes through another entry or a policy group, the result cannot be attributed to this connection.
-
Control On Demand
Go to Settings → On Demand and note its current status and matching conditions. If you need to control the connection manually while troubleshooting, temporarily disable On Demand to prevent automatic connections during network changes from affecting your observations. Restore the original setting after testing.
-
Restore rule-based routing
After verifying the connection, go to Home → Global Routing → Config and confirm the active configuration in Config. Test the same destination again. If it fails only in Config, first check which rule matches and where its policy points.
Updating a subscription may refresh an address or parameter changed by the service provider, but it will not remove a local UDP restriction. Note where the settings came from before editing so you can distinguish manual changes from the subscription's original values after an update. Subscription links contain access credentials; hide the full link, passwords, and authentication details when sharing diagnostic information.
6. Use Global Routing to distinguish connection issues from rule issues
Global Routing determines how traffic is routed; it does not speed up a particular protocol. When troubleshooting Hysteria2, controlled comparisons using different routing modes are more useful than changing DNS, rules, and protocol parameters all at once.
Config: follow the configuration
RecommendedRoutes traffic according to the active configuration's rules and policies. Whether a matching request uses Hysteria2 depends on the policy it points to and the selection actually made. For requests that match no earlier rule, check FINAL.
Best for: everyday rule-based routing after verifying the connection
Proxy: isolate rule effects
Switch overall routing to Proxy to see whether the test destination works through the selected proxy route. Still verify which entry is actually used; a policy group name alone does not prove which connection was selected.
Best for: temporarily verifying the Hysteria2 handshake and data transfer
Direct: establish a direct-connection baseline
Use Direct to check whether the destination is reachable over your local network path. Success only shows that the destination is reachable directly; it does not prove that the remote proxy server or its UDP port is working.
Best for: separating destination website problems from proxy-path problems
Scene in Global Routing applies scenario-based routing. Interpret the result in light of the configured scenes; it is not an unconditional baseline. To reproduce an issue, note the current routing mode and scene first, and make sure automatic switching does not send the two tests over different paths.
For example, the rules below show how to send example.com and its subdomains to a policy named PROXY, then fall back to DIRECT. The actual configuration must define that policy and select the entry being tested. DOMAIN-SUFFIX, GEOIP, and IP-CIDR match different conditions; FINAL provides the fallback. None of them changes Hysteria2's congestion control.
[Rule]
DOMAIN-SUFFIX,example.com,PROXY
FINAL,DIRECT
Bottom line: if it fails only in Config, check the route match first
Keep the destination and Hysteria2 service entry unchanged. If Proxy works but Config does not, first check whether the rules send the request to the expected policy, then check DNS and destination matching. Do not immediately re-enter the authentication details.
7. Troubleshoot connection timeouts, low throughput, and certificate errors
Latency tests may use different probing methods, and their results do not necessarily cover Hysteria2 authentication and data transfer end to end. Record the test type, exact error, network type, and current outbound entry. “Normal latency, but nothing loads” is only a starting point; it does not prove that the server or DNS is at fault.
- Repeated timeouts: First check the server address and destination UDP port, then inspect the local network path, server firewall, and service process. For example, a TCP 443 connection test cannot replace checking UDP 443.
- Certificate verification error: Check the system clock, certificate validity period, certificate chain, and SNI. If you connect by IP but the certificate is issued for a domain, use the verification name specified by the service provider. Do not disable certificate verification instead of fixing the problem.
- Authentication fails after the handshake: Check that the password is complete and has no leading or trailing spaces, and confirm that the server uses the matching authentication method. If obfuscation is enabled, check those parameters separately.
- Small requests work, but large transfers stall: Check for sustained packet loss, rate limits, path MTU issues, and UDP fragmentation problems. Do not guess a fixed MTU without evidence; narrow down the cause by switching networks and checking server logs.
- Transfers work, but domain names do not: Distinguish resolution of the server's domain from resolution of the destination website's domain. Check the current DNS path, rule results, and error logs rather than blaming every failure on QUIC.
Compare under consistent conditions, not with a single speed test
Use the same device, network, server, and test destination for every run. Record whether the connection establishes, time to first response, sustained throughput, and interactive latency during transfer. You could use the same 60-second observation window for each run and repeat it three times. This is a suggested test method, not product performance data.
- Run a test with the network idle and record baseline RTT and variation.
- Then run a sustained transfer and watch whether interactive latency rises as throughput increases.
- Change only one factor at a time, such as network access type or bandwidth target, and keep the rest unchanged.
- Restore the original Global Routing and On Demand settings, then confirm that everyday rules still work as expected.
Choose Hysteria2 based on repeatable performance in your own tasks. When UDP is available, the server details are correct, and the bandwidth target is reasonable, it may improve transfers on some high-loss or high-latency paths. If the issue is caused by UDP restrictions, server congestion, or incorrect routing, fix that part first rather than expecting the protocol name alone to resolve it.
Download Shadowrocket and continue setup
The iPhone and iPad version is a one-time $2.99 purchase on the App Store. The Mac (Apple silicon and Intel), Windows, Android, and Linux versions are free. Get the client for your device, then follow the guide to import your existing server details. For iPhone and iPad system requirements, see the App Store listing.