Reviewing the Results of Running AWS WAF Anti-DDoS Managed Rules on CloudFront for One Year
This page has been translated by machine translation. View original
Introduction
In February 2025, this site received a spike of around 50,000 requests per hour from ClaudeBot. Since then, traffic spikes from AI crawlers have continued to occur about once every one to two months. Judging by the Bot Control CategoryAI criterion, spikes exceeding 2σ have been recorded about eight times a year. I summarized the background in a previous article.
On June 12, 2025, the AWS WAF Anti-DDoS managed rule (AWSManagedRulesAntiDDoSRuleSet) became generally available (GA). Takaku's article covers the feature overview in detail.
As an additional safeguard against these recurring spikes, this site adopted the Anti-DDoS managed rule. This article reports the results of following the commonly given advice when adopting it, "start by observing in Count mode," exactly as stated, for one year.
This article checks the following four points.
- DDoS event detection record (did the rule fire in 13 months?)
- The reality of traffic spikes (did spikes actually occur in the first place?)
- Impact on the service during spikes (did the error rate worsen?)
- Operating cost (how much did it cost over 13 months?)
What We Verified
Checking the Web ACL Configuration
We placed the Anti-DDoS managed rule at Priority 100 and ran the entire rule group and all sub-rules in Count mode. The Challenge sensitivity (Sensitivity) is HIGH, and SensitivityToBlock is LOW. API paths and static assets are excluded with ExemptUriRegularExpressions.
Here is an excerpt of the relevant part of the configuration (account IDs and resource IDs are masked).
{
"WebACL": {
"Name": "my-waf-bot-webacl",
"ARN": "arn:aws:wafv2:us-east-1:123456789012:global/webacl/my-waf-bot-webacl/xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx",
"Rules": [
{
"Name": "AWS-AWSManagedRulesAntiDDoSRuleSet",
"Priority": 100,
"Statement": {
"ManagedRuleGroupStatement": {
"VendorName": "AWS",
"Name": "AWSManagedRulesAntiDDoSRuleSet",
"ManagedRuleGroupConfigs": [
{
"AWSManagedRulesAntiDDoSRuleSet": {
"ClientSideActionConfig": {
"Challenge": {
"UsageOfAction": "ENABLED",
"Sensitivity": "HIGH",
"ExemptUriRegularExpressions": [
{
"RegexString": "\\/api\\/|\\.(acc|avi|css|gif|ico|jpe?g|js|json|mp[34]|ogg|otf|pdf|png|tiff?|ttf|webm|webp|woff2?|xml)$"
}
]
}
},
"SensitivityToBlock": "LOW"
}
}
],
"RuleActionOverrides": [
{
"Name": "ChallengeAllDuringEvent",
"ActionToUse": { "Count": {} }
},
{
"Name": "ChallengeDDoSRequests",
"ActionToUse": { "Count": {} }
},
{
"Name": "DDoSRequests",
"ActionToUse": { "Count": {} }
}
]
}
},
"OverrideAction": { "Count": {} },
"VisibilityConfig": {
"SampledRequestsEnabled": true,
"CloudWatchMetricsEnabled": true,
"MetricName": "AWS-AWSManagedRulesAntiDDoSRuleSet"
}
}
]
}
}
The OverrideAction and the RuleActionOverrides of the three sub-rules (ChallengeAllDuringEvent / ChallengeDDoSRequests / DDoSRequests) are all set to Count, making this observation-only. The plan was to first check the firing status in Count mode, and switch to the actual actions once detection was confirmed. Specifically, you change OverrideAction from Count to None and remove the RuleActionOverrides of each sub-rule. This enables the default actions of each sub-rule, Challenge / Block.
DDoS Event Detection Results
Since it was operated in Count mode, if the rule had fired, data points would remain in the CloudWatch metrics. We checked the metrics for the entire operating period, from June 2025 around GA through July 18, 2026, for the rule as a whole and for each sub-rule.
| Metric target | Number of data points |
|---|---|
| AWS-AWSManagedRulesAntiDDoSRuleSet (entire rule) | 0 |
| ChallengeAllDuringEvent | 0 |
| ChallengeDDoSRequests | 0 |
| DDoSRequests | 0 |
The result was no data points for any sub-rule. Within the range of CloudWatch metrics we checked, no requests matching any of the three Anti-DDoS sub-rules were recorded over the 13 months. In particular, ChallengeAllDuringEvent, which matches challengeable requests during an event, was also zero, so we could not confirm any behavior indicating DDoS event detection.
On the other hand, AWSManagedIPDDoSList of the Amazon IP Reputation List, a separate managed rule group, reacts on a daily basis. The trend over the last eight days is as follows (July 18, 2026 is a partial value up to the time of data retrieval).
| Date | CountedRequests |
|---|---|
| 2026-07-11 | 46 |
| 2026-07-12 | 51 |
| 2026-07-13 | 169 |
| 2026-07-14 | 25 |
| 2026-07-15 | 30 |
| 2026-07-16 | 62 |
| 2026-07-17 | 26 |
| 2026-07-18 | 3 |
Looking at the completed days of July 11–17, the IP DDoS List reacts at a pace of roughly 20–170 requests per day, which shows that access from IP addresses that AWS has identified as being involved in DDoS activity arrives routinely. However, this is an evaluation based on IP reputation and is distinct from the "DDoS event detection" performed by the Anti-DDoS managed rule.
For reference, the sub-rule composition of the Anti-DDoS managed rule is as follows. All three sub-rules take effect after a DDoS event is detected, so they do not trigger unless an event is detected.
| Rule name | Action | Description |
|---|---|---|
| ChallengeAllDuringEvent | Challenge | Challenges challengeable requests during an event (other than excluded URIs) |
| ChallengeDDoSRequests | Challenge | Challenges requests with a suspicion level at or above the configured sensitivity |
| DDoSRequests | Block | Blocks requests with a suspicion level at or above the configured sensitivity |
Traffic Spike Analysis
Could it be that the Anti-DDoS rule did not fire simply because no spikes occurred in the first place? We check this with daily traffic statistics.
The aggregation period was 18 months (January 18, 2025 to July 18, 2026). The starting point is 18 months before the data retrieval date. By including the roughly five months before the Anti-DDoS adoption as a baseline, we can check for the presence of spikes before and after adoption.
| Metric | Value |
|---|---|
| Average daily requests | 1,597,924 |
| Median | 1,486,776 |
| Standard deviation | 841,473 |
| Minimum | 489,609 |
| Maximum | 6,010,556 |
| 2σ threshold | 3,280,870 |
Extracting the days that exceeded the threshold of mean + 2σ (about 3.28 million requests/day), the following four days qualified. Note that this is a 2σ exceedance based on overall traffic, and is a different metric from the "eight times a year" based on the Bot Control CategoryAI criterion mentioned in the "Introduction."
| Date | Requests | Ratio to average | Notes |
|---|---|---|---|
| 2026-04-09 | 6,010,556 | 3.8x | Largest spike |
| 2026-04-10 | 3,935,606 | 2.5x | Continuation from the previous day |
| 2025-12-02 | 3,792,110 | 2.4x | |
| 2025-12-03 | 3,474,911 | 2.2x |
Let's look in detail, hour by hour, at the largest spike on April 9–10, 2026. Note that the daily aggregation is based on UTC, while the hourly analysis is expressed in JST. Part of the increase that occurred on April 10 in JST is included in April 9 in the UTC-based daily table.
| Item | Value |
|---|---|
| Peak time | 2026-04-10 15:00 JST |
| Peak hourly value | 418,867 req |
| Pattern | In JST, traffic began increasing around 4/10 00:00 and peaked during the daytime. Not a sudden burst but an overall baseline-lifting pattern |
| CountedRequests (4/10) | 7,364,713 |
| AllowedRequests (4/10) | 3,935,606 |
| BlockedRequests (4/10) | 593 |
| IP DDoS List hits (4/10) | 45 |
| Anti-DDoS rule hits | Zero (did not fire) |
On April 10, CountedRequests reached 7.3 million, but the Anti-DDoS managed rule's response was zero. AllowedRequests was 3.93 million. Because Count is a non-terminating action and the same request is counted redundantly in the metrics of multiple rules, the two cannot simply be subtracted from each other. However, we can confirm that many matches occurred in Count rules such as Bot Control. IP DDoS List hits were also only 45 on this day. During this spike period, which includes April 9, the daily maximum (3.8x the average), the Anti-DDoS managed rule never fired.
Service Impact During Spikes
Here is TotalErrorRate around the largest spike (April 8–11, 2026).
| Date | TotalErrorRate |
|---|---|
| 2026-04-08 | 0.898% |
| 2026-04-09 | 0.556% |
| 2026-04-10 | 0.649% |
| 2026-04-11 | 0.865% |
The CloudFront 5xxErrorRate (the proportion of 5xx responses CloudFront returned to viewers) over this period was 0.000% for the spike days of April 9–10 (48 hourly periods). Even including the day when traffic reached 3.8 times the average, we could not confirm any increase in 5xx responses.
Next is around the December 2025 spike (December 1–5).
| Date | TotalErrorRate |
|---|---|
| 2025-12-01 | 2.186% |
| 2025-12-02 | 1.459% |
| 2025-12-03 | 0.955% |
| 2025-12-04 | 1.162% |
| 2025-12-05 | 1.160% |
The 5xx error rate over this period averaged 0.002556%. Even the peak was 0.1188% at 23:00 on December 4, and that was late at night after the spike had subsided. The spike peak and the 5xx peak did not coincide.
For comparison, let's also look at a normal period without spikes (July 14–17, 2026).
| Date | TotalErrorRate |
|---|---|
| 2026-07-14 | 2.974% |
| 2026-07-15 | 3.205% |
| 2026-07-16 | 3.689% |
| 2026-07-17 | 2.775% |
The 5xx error rate during the normal period averaged 0.000019%. TotalErrorRate is 2.7–3.7%, but since 5xxErrorRate is extremely small, the majority of it is 4xx.
Placing the three periods side by side, the TotalErrorRate during spikes (0.5–0.6%) turned out to be lower than during normal periods (2.7–3.7%). It is likely that during spikes the proportion of successful responses (2xx/3xx) increased relatively, and the increase in total requests lowered TotalErrorRate. At least judging by CloudFront's 5xxErrorRate, we could not confirm any increase in 5xx responses to viewers even during the period when traffic reached 3.8 times the average.
Cost Results
This is the monthly cost breakdown over the 13 months. The Anti-DDoS managed rule's charge is the sum of the subscription ($20/month, prorated hourly) and the request processing charge ($0.15 per million requests). This is an additional charge incurred separately from the basic WAF request charge ($0.60 per million requests). The cumulative total for the 13 billing months from June 2025 to June 2026 was $393.33, with a monthly average of $30.26 (July 2026 is not included because billing has not yet been finalized).
| Month | Subscription | Requests | Total | Notes |
|---|---|---|---|---|
| 2025-06 | — | $0.85 | $0.85 | Started mid-month (GA: 2025/6/12) |
| 2025-07 | $20.00 | $11.03 | $31.03 | |
| 2025-08 | $20.00 | $14.89 | $34.89 | |
| 2025-09 | $20.00 | $18.18 | $38.18 | |
| 2025-10 | $20.00 | $12.27 | $32.27 | |
| 2025-11 | $20.00 | $12.03 | $32.03 | |
| 2025-12 | $20.00 | $13.27 | $33.27 | |
| 2026-01 | $20.00 | $12.05 | $32.05 | |
| 2026-02 | $20.00 | $11.63 | $31.63 | |
| 2026-03 | $20.00 | $12.97 | $32.97 | |
| 2026-04 | $20.00 | $15.30 | $35.30 | |
| 2026-05 | $20.00 | $10.42 | $30.42 | |
| 2026-06 | $20.00 | $8.44 | $28.44 |
In the first month (June 2025), no subscription fee was billed, and only the request charge of $0.85 was recorded. Officially, the subscription fee is said to be prorated hourly, but we could not determine from the billing data why it was not recorded in the first month. From then on, the $20 subscription has been incurred every month. Details of the pricing structure are described on the AWS WAF pricing page. For reference, AWS Shield Advanced, which likewise strengthens DDoS protection, costs $3,000/month. However, the scope of features differs greatly, for example WAF charges and DRT support are included, so a simple price comparison is not possible.
Using the challengeable-request Label
Although DDoS event detection was zero, this rule has a secondary function that we used continuously: the awswaf:managed:aws:anti-ddos:challengeable-request label.
Of the six labels that the Anti-DDoS managed rule applies, only challengeable-request is applied at all times regardless of whether a DDoS event is occurring (other than excluded URIs).
| Label | Description | Application condition |
|---|---|---|
| awswaf:managed:aws:anti-ddos:event-detected | DDoS event detected | Only during an event |
| awswaf:managed:aws:anti-ddos:ddos-request | Request from a source suspected of participating in DDoS | Only during an event |
| awswaf:managed:aws:anti-ddos:low-suspicion-ddos-request | Possibly a DDoS attack request | Only during an event |
| awswaf:managed:aws:anti-ddos:medium-suspicion-ddos-request | Likely a DDoS attack request | Only during an event |
| awswaf:managed:aws:anti-ddos:high-suspicion-ddos-request | Very likely a DDoS attack request | Only during an event |
| awswaf:managed:aws:anti-ddos:challengeable-request | Request that can be challenged (other than excluded URIs) | Always applied |
On this site, we used this always-applied challengeable-request label as a condition in a Rate-based rule. Specifically, it is incorporated into the scope-down statement of the waf-bot-Ratebased-JA4-NonJPUS rule at Priority 203. Even with zero DDoS event detections, the use of this label alone made it worthwhile to keep the Anti-DDoS managed rule.
Why Didn't It Fire?
The specific reason it did not fire cannot be determined from public information and the metrics we obtained this time. The detection condition of the Anti-DDoS managed rule is a "significant deviation" from the traffic baseline. This baseline is built automatically by AWS, and users have no means of directly adjusting the detection threshold.
The peak hourly value was 418,867 requests (about 116 req/sec on an hourly average). What we were able to confirm this time is that no Anti-DDoS sub-rule matches were recorded for this traffic change. Whether CloudFront could handle the traffic and whether the Anti-DDoS detection conditions are met are separate matters.
Details are described in the official documentation on the Anti-DDoS rule group and the documentation on application layer DDoS protection.
Summary
After running the Anti-DDoS managed rule in Count mode for 13 months in a CloudFront environment, the metrics indicating DDoS event detection were zero. Even including a spike reaching 3.8 times the average, no Anti-DDoS sub-rule matches were recorded, and no increase in CloudFront 5xxErrorRate was seen during the spike periods we checked. The operational result is that, at this site's traffic scale and within the 13-month observation range, the rule did not trigger on spikes of the magnitude we observed. On the other hand, being able to use the always-applied challengeable-request label as a condition in a Rate-based rule was a secondary benefit. Whether you can pay around $30 a month as "insurance, even if it never fires" depends on the scale of your site and your risk tolerance. For small sites, where the $20/month subscription is a large proportion of the cost of the workload being protected, it may be wise to review cost-effectiveness periodically. For this site, we decided to keep it in Count mode, for the purposes of detection when a DDoS event occurs and use of the challengeable-request label.

